diff --git "a/openvino_model.xml" "b/openvino_model.xml"
--- "a/openvino_model.xml"
+++ "b/openvino_model.xml"
@@ -19,59 +19,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="2" name="Constant_1059003" type="Const" version="opset1">
-			<data element_type="i8" shape="250002, 768" offset="0" size="192001536" />
+		<layer id="2" name="self.roberta.embeddings.word_embeddings.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="250002, 768" offset="0" size="192001536" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>250002</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="3" name="Convert_1059004" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="3" name="Convert_16472" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>250002</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>250002</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="4" name="Constant_1059005" type="Const" version="opset1">
-			<data element_type="f32" shape="250002, 1" offset="192001536" size="1000008" />
+		<layer id="4" name="self.roberta.embeddings.word_embeddings.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="250002, 1" offset="192001536" size="250002" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>250002</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="5" name="Convert_16475" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>250002</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>250002</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="5" name="__module.roberta.embeddings.word_embeddings/aten::embedding/Gather/fq_weights_0" type="Multiply" version="opset1">
+		<layer id="6" name="self.roberta.embeddings.word_embeddings.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>250002</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>250002</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>250002</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="7" name="self.roberta.embeddings.word_embeddings.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="250002, 1" offset="192251538" size="500004" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>250002</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="8" name="self.roberta.embeddings.word_embeddings.weight/fq_weights_0" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>250002</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>250002</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>250002</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="9" name="self.roberta.embeddings.word_embeddings.weight/fq_weights_0/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>250002</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>250002</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="6" name="__module.roberta.embeddings.word_embeddings/aten::embedding/Convert" type="Convert" version="opset1">
+		<layer id="10" name="__module.roberta.embeddings.word_embeddings/aten::embedding/Convert" type="Convert" version="opset1">
 			<data destination_type="i32" />
 			<input>
 				<port id="0" precision="I64">
@@ -86,13 +144,13 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="7" name="__module.roberta.embeddings.word_embeddings/aten::embedding/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="" offset="193001544" size="4" />
+		<layer id="11" name="__module.roberta.embeddings.word_embeddings/aten::embedding/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="" offset="192751542" size="4" />
 			<output>
 				<port id="0" precision="I32" />
 			</output>
 		</layer>
-		<layer id="8" name="__module.roberta.embeddings.word_embeddings/aten::embedding/Gather" type="Gather" version="opset8">
+		<layer id="12" name="__module.roberta.embeddings.word_embeddings/aten::embedding/Gather" type="Gather" version="opset8">
 			<data batch_dims="0" />
 			<input>
 				<port id="0" precision="FP32">
@@ -106,83 +164,141 @@
 				<port id="2" precision="I32" />
 			</input>
 			<output>
-				<port id="3" precision="FP32" names="54,inputs_embeds">
+				<port id="3" precision="FP32" names="55,inputs_embeds">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="9" name="Constant_1059007" type="Const" version="opset1">
-			<data element_type="i8" shape="1, 768" offset="193001548" size="768" />
+		<layer id="13" name="self.roberta.embeddings.token_type_embeddings.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="1, 768" offset="192751546" size="768" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="10" name="Convert_1059008" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="14" name="Convert_311595" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="11" name="Constant_1059009" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1" offset="193002316" size="4" />
+		<layer id="15" name="self.roberta.embeddings.token_type_embeddings.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="1, 1" offset="192752314" size="1" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>1</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="16" name="Convert_311598" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>1</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="12" name="__module.roberta.embeddings.token_type_embeddings/aten::embedding/Gather/fq_weights_0" type="Multiply" version="opset1">
+		<layer id="17" name="self.roberta.embeddings.token_type_embeddings.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="18" name="self.roberta.embeddings.token_type_embeddings.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="1, 1" offset="192752315" size="2" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>1</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="19" name="self.roberta.embeddings.token_type_embeddings.weight/fq_weights_0" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>1</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="20" name="self.roberta.embeddings.token_type_embeddings.weight/fq_weights_0/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="13" name="__module.roberta/aten::slice/Slice" type="Const" version="opset1">
-			<data element_type="i64" shape="1, 514" offset="193002320" size="4112" />
+		<layer id="21" name="__module.roberta/aten::slice/Slice" type="Const" version="opset1">
+			<data element_type="i64" shape="1, 514" offset="192752317" size="4112" />
 			<output>
-				<port id="0" precision="I64" names="37">
+				<port id="0" precision="I64" names="38">
 					<dim>1</dim>
 					<dim>514</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="14" name="Constant_54" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="193006432" size="8" />
+		<layer id="22" name="Constant_56" type="Const" version="opset1">
+			<data element_type="i64" shape="1" offset="192756429" size="8" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="15" name="ShapeOf_9077" type="ShapeOf" version="opset3">
+		<layer id="23" name="ShapeOf_9141" type="ShapeOf" version="opset3">
 			<data output_type="i64" />
 			<input>
 				<port id="0" precision="I64">
@@ -196,21 +312,21 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="16" name="Constant_9285" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="193006440" size="8" />
+		<layer id="24" name="Constant_9409" type="Const" version="opset1">
+			<data element_type="i64" shape="1" offset="192756437" size="8" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="17" name="Constant_9079" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="193006432" size="8" />
+		<layer id="25" name="Constant_9143" type="Const" version="opset1">
+			<data element_type="i64" shape="" offset="192756429" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="18" name="Gather_9080" type="Gather" version="opset8">
+		<layer id="26" name="Gather_9144" type="Gather" version="opset8">
 			<data batch_dims="0" />
 			<input>
 				<port id="0" precision="I64">
@@ -222,28 +338,28 @@
 				<port id="2" precision="I64" />
 			</input>
 			<output>
-				<port id="3" precision="I64" names="36,67">
+				<port id="3" precision="I64" names="37,68">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="19" name="Constant_57" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="193006440" size="8" />
+		<layer id="27" name="Constant_59" type="Const" version="opset1">
+			<data element_type="i64" shape="1" offset="192756437" size="8" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="20" name="Constant_52" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="193006440" size="8" />
+		<layer id="28" name="Constant_54" type="Const" version="opset1">
+			<data element_type="i64" shape="1" offset="192756437" size="8" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="21" name="__module.roberta/aten::slice/Slice_1" type="Slice" version="opset8">
+		<layer id="29" name="__module.roberta/aten::slice/Slice_1" type="Slice" version="opset8">
 			<input>
 				<port id="0" precision="I64">
 					<dim>1</dim>
@@ -263,13 +379,13 @@
 				</port>
 			</input>
 			<output>
-				<port id="5" precision="I64" names="38,buffered_token_type_ids">
+				<port id="5" precision="I64" names="39,buffered_token_type_ids">
 					<dim>1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="22" name="__module.roberta/aten::expand/Broadcast" type="Broadcast" version="opset3">
+		<layer id="30" name="__module.roberta/aten::expand/Broadcast" type="Broadcast" version="opset3">
 			<data mode="bidirectional" />
 			<input>
 				<port id="0" precision="I64">
@@ -281,13 +397,13 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="I64" names="40">
+				<port id="2" precision="I64" names="41,input.1">
 					<dim>-1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="23" name="__module.roberta.embeddings.token_type_embeddings/aten::embedding/Convert" type="Convert" version="opset1">
+		<layer id="31" name="__module.roberta.embeddings.token_type_embeddings/aten::embedding/Convert" type="Convert" version="opset1">
 			<data destination_type="i32" />
 			<input>
 				<port id="0" precision="I64">
@@ -302,13 +418,13 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="24" name="__module.roberta.embeddings.token_type_embeddings/aten::embedding/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="" offset="193001544" size="4" />
+		<layer id="32" name="__module.roberta.embeddings.token_type_embeddings/aten::embedding/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="" offset="192751542" size="4" />
 			<output>
 				<port id="0" precision="I32" />
 			</output>
 		</layer>
-		<layer id="25" name="__module.roberta.embeddings.token_type_embeddings/aten::embedding/Gather" type="Gather" version="opset8">
+		<layer id="33" name="__module.roberta.embeddings.token_type_embeddings/aten::embedding/Gather" type="Gather" version="opset8">
 			<data batch_dims="0" />
 			<input>
 				<port id="0" precision="FP32">
@@ -322,14 +438,14 @@
 				<port id="2" precision="I32" />
 			</input>
 			<output>
-				<port id="3" precision="FP32" names="56,token_type_embeddings.1">
+				<port id="3" precision="FP32" names="57,token_type_embeddings.1">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="26" name="__module.roberta.embeddings/aten::add/Add" type="Add" version="opset1">
+		<layer id="34" name="__module.roberta.embeddings/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -344,67 +460,125 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="57_1">
+				<port id="2" precision="FP32" names="58_1">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="27" name="Constant_1059011" type="Const" version="opset1">
-			<data element_type="i8" shape="514, 768" offset="193006448" size="394752" />
+		<layer id="35" name="self.roberta.embeddings.position_embeddings.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="514, 768" offset="192756445" size="394752" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>514</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="28" name="Convert_1059012" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="36" name="Convert_311584" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>514</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>514</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="29" name="Constant_1059013" type="Const" version="opset1">
-			<data element_type="f32" shape="514, 1" offset="193401200" size="2056" />
+		<layer id="37" name="self.roberta.embeddings.position_embeddings.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="514, 1" offset="193151197" size="514" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>514</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="38" name="Convert_311587" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>514</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>514</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="30" name="__module.roberta.embeddings.position_embeddings/aten::embedding/Gather/fq_weights_0" type="Multiply" version="opset1">
+		<layer id="39" name="self.roberta.embeddings.position_embeddings.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>514</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>514</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>514</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="40" name="self.roberta.embeddings.position_embeddings.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="514, 1" offset="193151711" size="1028" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>514</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="41" name="self.roberta.embeddings.position_embeddings.weight/fq_weights_0" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>514</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>514</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>514</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="42" name="self.roberta.embeddings.position_embeddings.weight/fq_weights_0/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>514</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>514</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="31" name="Constant_8854" type="Const" version="opset1">
-			<data element_type="i64" shape="1, 1" offset="193006440" size="8" />
+		<layer id="43" name="Constant_8895" type="Const" version="opset1">
+			<data element_type="i64" shape="1, 1" offset="192756437" size="8" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>1</dim>
@@ -412,7 +586,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="32" name="__module.roberta.embeddings/aten::ne/NotEqual" type="NotEqual" version="opset1">
+		<layer id="44" name="__module.roberta.embeddings/aten::ne/NotEqual" type="NotEqual" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="I64">
@@ -425,13 +599,13 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="BOOL" names="45">
+				<port id="2" precision="BOOL" names="46">
 					<dim>-1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="33" name="__module.roberta.embeddings/aten::to/Convert" type="Convert" version="opset1">
+		<layer id="45" name="__module.roberta.embeddings/aten::to/Convert" type="Convert" version="opset1">
 			<data destination_type="i32" />
 			<input>
 				<port id="0" precision="BOOL">
@@ -440,19 +614,19 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="I32" names="46,mask">
+				<port id="1" precision="I32" names="47,mask">
 					<dim>-1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="34" name="29" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="193006440" size="8" />
+		<layer id="46" name="30" type="Const" version="opset1">
+			<data element_type="i64" shape="" offset="192756437" size="8" />
 			<output>
-				<port id="0" precision="I64" names="29" />
+				<port id="0" precision="I64" names="30" />
 			</output>
 		</layer>
-		<layer id="35" name="__module.roberta.embeddings/aten::cumsum/CumSum" type="CumSum" version="opset3">
+		<layer id="47" name="__module.roberta.embeddings/aten::cumsum/CumSum" type="CumSum" version="opset3">
 			<data exclusive="false" reverse="false" />
 			<input>
 				<port id="0" precision="I32">
@@ -462,13 +636,13 @@
 				<port id="1" precision="I64" />
 			</input>
 			<output>
-				<port id="2" precision="I32" names="47,48,49">
+				<port id="2" precision="I32" names="48,49,50">
 					<dim>-1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="36" name="__module.roberta.embeddings/aten::mul/Multiply" type="Multiply" version="opset1">
+		<layer id="48" name="__module.roberta.embeddings/aten::mul/Multiply" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="I32">
@@ -481,13 +655,13 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="I32" names="50,incremental_indices">
+				<port id="2" precision="I32" names="51,incremental_indices">
 					<dim>-1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="37" name="__module.roberta.embeddings/aten::to/Convert_1" type="Convert" version="opset1">
+		<layer id="49" name="__module.roberta.embeddings/aten::to/Convert_1" type="Convert" version="opset1">
 			<data destination_type="i64" />
 			<input>
 				<port id="0" precision="I32">
@@ -496,14 +670,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="I64" names="51">
+				<port id="1" precision="I64" names="52">
 					<dim>-1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="38" name="Constant_8855" type="Const" version="opset1">
-			<data element_type="i64" shape="1, 1" offset="193006440" size="8" />
+		<layer id="50" name="Constant_8896" type="Const" version="opset1">
+			<data element_type="i64" shape="1, 1" offset="192756437" size="8" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>1</dim>
@@ -511,7 +685,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="39" name="__module.roberta.embeddings/aten::add/Add_2" type="Add" version="opset1">
+		<layer id="51" name="__module.roberta.embeddings/aten::add/Add_2" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="I64">
@@ -524,13 +698,13 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="I64" names="52">
+				<port id="2" precision="I64" names="53,input.3">
 					<dim>-1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="40" name="__module.roberta.embeddings.position_embeddings/aten::embedding/Convert" type="Convert" version="opset1">
+		<layer id="52" name="__module.roberta.embeddings.position_embeddings/aten::embedding/Convert" type="Convert" version="opset1">
 			<data destination_type="i32" />
 			<input>
 				<port id="0" precision="I64">
@@ -545,13 +719,13 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="41" name="__module.roberta.embeddings.position_embeddings/aten::embedding/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="" offset="193001544" size="4" />
+		<layer id="53" name="__module.roberta.embeddings.position_embeddings/aten::embedding/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="" offset="192751542" size="4" />
 			<output>
 				<port id="0" precision="I32" />
 			</output>
 		</layer>
-		<layer id="42" name="__module.roberta.embeddings.position_embeddings/aten::embedding/Gather" type="Gather" version="opset8">
+		<layer id="54" name="__module.roberta.embeddings.position_embeddings/aten::embedding/Gather" type="Gather" version="opset8">
 			<data batch_dims="0" />
 			<input>
 				<port id="0" precision="FP32">
@@ -565,14 +739,14 @@
 				<port id="2" precision="I32" />
 			</input>
 			<output>
-				<port id="3" precision="FP32" names="59,position_embeddings.1">
+				<port id="3" precision="FP32" names="60,position_embeddings.1">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="43" name="__module.roberta.embeddings/aten::add_/Add" type="Add" version="opset1">
+		<layer id="55" name="__module.roberta.embeddings/aten::add_/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -587,22 +761,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="57,embeddings.1">
+				<port id="2" precision="FP32" names="58,embeddings.1">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="44" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="56" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="45" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="57" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -622,8 +796,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="46" name="Constant_8856" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="193403260" size="3072" />
+		<layer id="58" name="Constant_8897" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="193152743" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -632,7 +806,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="47" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="59" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -654,8 +828,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="48" name="Constant_8857" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="193406332" size="3072" />
+		<layer id="60" name="Constant_8898" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="193155815" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -664,7 +838,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="49" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="61" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -679,143 +853,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="64,input.1">
+				<port id="2" precision="FP32" names="65,input.7">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="50" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="193409404" size="3072" />
+		<layer id="62" name="self.roberta.encoder.layer.0.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="193158887" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="51" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="63" name="Convert_124148" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="52" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="193412476" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="53" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="193412480" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="54" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="193412476" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="55" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="193412480" size="4" />
+		<layer id="64" name="self.roberta.encoder.layer.0.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="193748711" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="56" name="__module.roberta.embeddings.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="65" name="Convert_124151" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="57" name="Constant_1059015" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="193412484" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="66" name="self.roberta.encoder.layer.0.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="58" name="Convert_1059016" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="59" name="Constant_1059017" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="194002308" size="3072" />
+		<layer id="67" name="self.roberta.encoder.layer.0.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="193749479" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="60" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="68" name="self.roberta.encoder.layer.0.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="69" name="self.roberta.encoder.layer.0.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="61" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="70" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -836,8 +991,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="62" name="Constant_8858" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="194005380" size="3072" />
+		<layer id="71" name="Constant_8899" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="193751015" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -846,7 +1001,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="63" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="72" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -861,67 +1016,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="114,x.1">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="64" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="194008452" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="65" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="194008456" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="66" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="194008452" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="67" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="194008456" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="68" name="__module.roberta.encoder.layer.0.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="117">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="69" name="__module.roberta.encoder.layer.0.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="73" name="Constant_9257" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="70" name="__module.roberta.encoder.layer.0.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="74" name="__module.roberta.encoder.layer.0.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -934,7 +1044,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="118,x.3">
+				<port id="2" precision="FP32" names="119">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -942,15 +1052,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="71" name="Constant_273" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="75" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="119">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="72" name="__module.roberta.encoder.layer.0.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="76" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -958,12 +1068,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="120">
+				<port id="2" precision="FP32" names="120,query_layer.1">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -971,91 +1081,118 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="73" name="Constant_1059019" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="194008524" size="589824" />
+		<layer id="77" name="self.roberta.encoder.layer.0.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="193754135" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="74" name="Convert_1059020" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="78" name="Convert_120160" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="75" name="Constant_1059021" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="194598348" size="3072" />
+		<layer id="79" name="self.roberta.encoder.layer.0.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="194343959" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="80" name="Convert_120163" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="76" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="81" name="self.roberta.encoder.layer.0.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="77" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+		<layer id="82" name="self.roberta.encoder.layer.0.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="194344727" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32">
+			</output>
+		</layer>
+		<layer id="83" name="self.roberta.encoder.layer.0.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="78" name="Constant_8859" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="194601420" size="3072" />
+		<layer id="84" name="self.roberta.encoder.layer.0.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="79" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/Add" type="Add" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="85" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -1063,73 +1200,59 @@
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="123,x.5">
+				<port id="2" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="80" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="194604492" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="81" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="194604496" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="82" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="194604492" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="83" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="194604496" size="4" />
+		<layer id="86" name="Constant_8900" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="194346263" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="84" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="87" name="__module.roberta.encoder.layer.0.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="123">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="85" name="__module.roberta.encoder.layer.0.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="88" name="Constant_9258" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="86" name="__module.roberta.encoder.layer.0.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="89" name="__module.roberta.encoder.layer.0.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1142,7 +1265,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="127,x.7">
+				<port id="2" precision="FP32" names="125">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -1150,15 +1273,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="87" name="Constant_307" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="90" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="128">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="88" name="__module.roberta.encoder.layer.0.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="91" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -1166,12 +1289,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="129">
+				<port id="2" precision="FP32" names="126,key_layer.1">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -1179,59 +1302,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="89" name="Constant_1059023" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="194604500" size="589824" />
+		<layer id="92" name="self.roberta.encoder.layer.0.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="194349335" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="90" name="Convert_1059024" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="93" name="Convert_116172" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="91" name="Constant_1059025" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="195194324" size="3072" />
+		<layer id="94" name="self.roberta.encoder.layer.0.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="194939159" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="95" name="Convert_116175" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="92" name="__module.roberta.encoder.layer.0.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="96" name="self.roberta.encoder.layer.0.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="97" name="self.roberta.encoder.layer.0.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="194939927" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="98" name="self.roberta.encoder.layer.0.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="99" name="self.roberta.encoder.layer.0.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="93" name="__module.roberta.encoder.layer.0.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="100" name="__module.roberta.encoder.layer.0.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1252,8 +1433,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="94" name="Constant_8860" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="195197396" size="3072" />
+		<layer id="101" name="Constant_8901" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="194941463" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -1262,7 +1443,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="95" name="__module.roberta.encoder.layer.0.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="102" name="__module.roberta.encoder.layer.0.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1277,22 +1458,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="132,x.9">
+				<port id="2" precision="FP32" names="129">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="96" name="__module.roberta.encoder.layer.0.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="103" name="Constant_9259" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="97" name="__module.roberta.encoder.layer.0.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="104" name="__module.roberta.encoder.layer.0.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1305,7 +1486,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="136,x.11">
+				<port id="2" precision="FP32" names="131">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -1313,15 +1494,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="98" name="Constant_341" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="105" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="137">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="99" name="__module.roberta.encoder.layer.0.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="106" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -1329,12 +1510,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="138">
+				<port id="2" precision="FP32" names="132,value_layer.1">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -1342,8 +1523,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="100" name="Constant_8862" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 1, 1" offset="195200468" size="4" />
+		<layer id="107" name="Constant_8902" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 1, 1" offset="194944535" size="4" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -1353,7 +1534,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="101" name="__module.roberta/aten::unsqueeze/Unsqueeze" type="Unsqueeze" version="opset1">
+		<layer id="108" name="__module.roberta/aten::unsqueeze/Unsqueeze" type="Unsqueeze" version="opset1">
 			<input>
 				<port id="0" precision="I64">
 					<dim>-1</dim>
@@ -1362,20 +1543,20 @@
 				<port id="1" precision="I64" />
 			</input>
 			<output>
-				<port id="2" precision="I64" names="69">
+				<port id="2" precision="I64" names="70">
 					<dim>-1</dim>
 					<dim>1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="102" name="16" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="195200472" size="8" />
+		<layer id="109" name="17" type="Const" version="opset1">
+			<data element_type="i64" shape="" offset="194944539" size="8" />
 			<output>
-				<port id="0" precision="I64" names="16" />
+				<port id="0" precision="I64" names="17" />
 			</output>
 		</layer>
-		<layer id="103" name="__module.roberta/aten::unsqueeze/Unsqueeze_1" type="Unsqueeze" version="opset1">
+		<layer id="110" name="__module.roberta/aten::unsqueeze/Unsqueeze_1" type="Unsqueeze" version="opset1">
 			<input>
 				<port id="0" precision="I64">
 					<dim>-1</dim>
@@ -1385,7 +1566,7 @@
 				<port id="1" precision="I64" />
 			</input>
 			<output>
-				<port id="2" precision="I64" names="70,71">
+				<port id="2" precision="I64" names="71,72">
 					<dim>-1</dim>
 					<dim>1</dim>
 					<dim>1</dim>
@@ -1393,21 +1574,21 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="104" name="Constant_9084" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="193006432" size="8" />
+		<layer id="111" name="Constant_9148" type="Const" version="opset1">
+			<data element_type="i64" shape="1" offset="192756429" size="8" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="105" name="Constant_9085" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="193006432" size="8" />
+		<layer id="112" name="Constant_9149" type="Const" version="opset1">
+			<data element_type="i64" shape="" offset="192756429" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="106" name="Gather_9086" type="Gather" version="opset8">
+		<layer id="113" name="Gather_9150" type="Gather" version="opset8">
 			<data batch_dims="0" />
 			<input>
 				<port id="0" precision="I64">
@@ -1419,34 +1600,34 @@
 				<port id="2" precision="I64" />
 			</input>
 			<output>
-				<port id="3" precision="I64" names="35,66">
+				<port id="3" precision="I64" names="113,172,231,290,349,36,408,467,526,585,644,67,703,762">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="107" name="Constant_7643" type="Const" version="opset1">
-			<data element_type="i64" shape="1" offset="193006440" size="8" />
+		<layer id="114" name="Constant_7873" type="Const" version="opset1">
+			<data element_type="i64" shape="1" offset="192756437" size="8" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="108" name="Constant_9287" type="Const" version="opset1">
-			<data element_type="i64" shape="2" offset="195200480" size="16" />
+		<layer id="115" name="Constant_9411" type="Const" version="opset1">
+			<data element_type="i64" shape="2" offset="194944547" size="16" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>2</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="109" name="Constant_9288" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="193006432" size="8" />
+		<layer id="116" name="Constant_9412" type="Const" version="opset1">
+			<data element_type="i64" shape="" offset="192756429" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="110" name="Gather_9289" type="Gather" version="opset8">
+		<layer id="117" name="Gather_9413" type="Gather" version="opset8">
 			<data batch_dims="0" />
 			<input>
 				<port id="0" precision="I64">
@@ -1463,7 +1644,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="111" name="__module.roberta/prim::ListConstruct_1" type="Concat" version="opset1">
+		<layer id="118" name="__module.roberta/prim::ListConstruct_1" type="Concat" version="opset1">
 			<data axis="0" />
 			<input>
 				<port id="0" precision="I64">
@@ -1482,7 +1663,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="112" name="__module.roberta/aten::expand/Broadcast_1" type="Broadcast" version="opset3">
+		<layer id="119" name="__module.roberta/aten::expand/Broadcast_1" type="Broadcast" version="opset3">
 			<data mode="bidirectional" />
 			<input>
 				<port id="0" precision="I64">
@@ -1496,7 +1677,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="I64" names="73">
+				<port id="2" precision="I64" names="74">
 					<dim>-1</dim>
 					<dim>1</dim>
 					<dim>-1</dim>
@@ -1504,7 +1685,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="113" name="__module.roberta/aten::to/Convert" type="Convert" version="opset1">
+		<layer id="120" name="__module.roberta/aten::to/Convert_1" type="Convert" version="opset1">
 			<data destination_type="f32" />
 			<input>
 				<port id="0" precision="I64">
@@ -1515,43 +1696,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="74">
-					<dim>-1</dim>
-					<dim>1</dim>
-					<dim>-1</dim>
-					<dim>-1</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="114" name="Constant_8861" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 1, 1" offset="195200468" size="4" />
-			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="115" name="__module.roberta/aten::rsub/Multiply" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>1</dim>
-					<dim>-1</dim>
-					<dim>-1</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-					<dim>1</dim>
-				</port>
-			</input>
-			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32" names="75,expanded_mask">
 					<dim>-1</dim>
 					<dim>1</dim>
 					<dim>-1</dim>
@@ -1559,7 +1704,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="116" name="__module.roberta/aten::rsub/Subtract" type="Subtract" version="opset1">
+		<layer id="121" name="__module.roberta/aten::sub/Subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1576,7 +1721,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="75,inverted_mask">
+				<port id="2" precision="FP32" names="78,inverted_mask">
 					<dim>-1</dim>
 					<dim>1</dim>
 					<dim>-1</dim>
@@ -1584,7 +1729,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="117" name="__module.roberta/aten::to/Convert_1" type="Convert" version="opset1">
+		<layer id="122" name="__module.roberta/aten::to/Convert_2" type="Convert" version="opset1">
 			<data destination_type="boolean" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1595,7 +1740,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="BOOL" names="76">
+				<port id="1" precision="BOOL" names="79">
 					<dim>-1</dim>
 					<dim>1</dim>
 					<dim>-1</dim>
@@ -1603,13 +1748,13 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="118" name="__module.roberta/aten::masked_fill/ConvertLike" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="195200496" size="4" />
+		<layer id="123" name="__module.roberta/aten::masked_fill/ConvertLike" type="Const" version="opset1">
+			<data element_type="f32" shape="" offset="194944563" size="4" />
 			<output>
 				<port id="0" precision="FP32" />
 			</output>
 		</layer>
-		<layer id="119" name="__module.roberta/aten::masked_fill/Select" type="Select" version="opset1">
+		<layer id="124" name="__module.roberta/aten::masked_fill/Select" type="Select" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="BOOL">
@@ -1627,7 +1772,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="3" precision="FP32" names="77">
+				<port id="3" precision="FP32" names="80,attention_mask_1">
 					<dim>-1</dim>
 					<dim>1</dim>
 					<dim>-1</dim>
@@ -1635,7 +1780,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="120" name="__module.roberta.encoder.layer.0.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="125" name="__module.roberta.encoder.layer.0.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1664,7 +1809,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="139,attn_output.1">
+				<port id="4" precision="FP32" names="133,attn_output.1">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -1672,15 +1817,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="121" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="126" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="122" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="127" name="__module.roberta.encoder.layer.0.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -1693,7 +1838,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="140,attn_output.3">
+				<port id="2" precision="FP32" names="134,attn_output.3">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -1701,15 +1846,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="123" name="Constant_9169" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="128" name="Constant_9260" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="124" name="__module.roberta.encoder.layer.0.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="129" name="__module.roberta.encoder.layer.0.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1723,143 +1868,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="142">
+				<port id="2" precision="FP32" names="136,input.9">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="125" name="__module.roberta.encoder.layer.0.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="195200540" size="3072" />
+		<layer id="130" name="self.roberta.encoder.layer.0.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="194944591" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="126" name="__module.roberta.encoder.layer.0.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="131" name="Convert_128136" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="127" name="__module.roberta.encoder.layer.0.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="195203612" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="128" name="__module.roberta.encoder.layer.0.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="195203616" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="129" name="__module.roberta.encoder.layer.0.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="195203612" size="4" />
+		<layer id="132" name="self.roberta.encoder.layer.0.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="195534415" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="130" name="__module.roberta.encoder.layer.0.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="195203616" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="131" name="__module.roberta.encoder.layer.0.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="133" name="Convert_128139" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="132" name="Constant_1059027" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="195203620" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="134" name="self.roberta.encoder.layer.0.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="133" name="Convert_1059028" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="134" name="Constant_1059029" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="195793444" size="3072" />
+		<layer id="135" name="self.roberta.encoder.layer.0.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="195535183" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="135" name="__module.roberta.encoder.layer.0.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="136" name="self.roberta.encoder.layer.0.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="137" name="self.roberta.encoder.layer.0.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="136" name="__module.roberta.encoder.layer.0.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="138" name="__module.roberta.encoder.layer.0.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1880,8 +2006,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="137" name="Constant_8863" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="195796516" size="3072" />
+		<layer id="139" name="Constant_8903" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="195536719" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -1890,7 +2016,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="138" name="__module.roberta.encoder.layer.0.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="140" name="__module.roberta.encoder.layer.0.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1905,14 +2031,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="147,input.3">
+				<port id="2" precision="FP32" names="141,input.11">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="139" name="__module.roberta.encoder.layer.0.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="141" name="__module.roberta.encoder.layer.0.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1927,22 +2053,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="149">
+				<port id="2" precision="FP32" names="143,input.13">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="140" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="142" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="141" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="143" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1962,8 +2088,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="142" name="Constant_8864" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="195799588" size="3072" />
+		<layer id="144" name="Constant_8904" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="195539791" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -1972,7 +2098,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="143" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="145" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -1994,8 +2120,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="144" name="Constant_8865" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="195802660" size="3072" />
+		<layer id="146" name="Constant_8905" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="195542863" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -2004,7 +2130,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="145" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="147" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2019,143 +2145,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="153,input_tensor.1">
+				<port id="2" precision="FP32" names="147,input.15">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="146" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="195805732" size="3072" />
+		<layer id="148" name="self.roberta.encoder.layer.0.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="195545935" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="147" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="149" name="Convert_20460" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="148" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="195808804" size="4" />
+		<layer id="150" name="self.roberta.encoder.layer.0.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="197905231" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="149" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="195808808" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="150" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="195808804" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="151" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="195808808" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="152" name="__module.roberta.encoder.layer.0.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="151" name="Convert_20463" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="153" name="Constant_1059031" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="195808812" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="154" name="Convert_1059032" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="152" name="self.roberta.encoder.layer.0.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="155" name="Constant_1059033" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="198168108" size="12288" />
+		<layer id="153" name="self.roberta.encoder.layer.0.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="197908303" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="156" name="__module.roberta.encoder.layer.0.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="154" name="self.roberta.encoder.layer.0.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="155" name="self.roberta.encoder.layer.0.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="157" name="__module.roberta.encoder.layer.0.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="156" name="__module.roberta.encoder.layer.0.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2176,8 +2283,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="158" name="Constant_8866" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="198180396" size="12288" />
+		<layer id="157" name="Constant_8906" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="197914447" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -2186,7 +2293,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="159" name="__module.roberta.encoder.layer.0.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="158" name="__module.roberta.encoder.layer.0.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2201,14 +2308,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="157">
+				<port id="2" precision="FP32" names="151,input.17">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="160" name="__module.roberta.encoder.layer.0.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="159" name="__module.roberta.encoder.layer.0.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2218,143 +2325,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="158">
+				<port id="1" precision="FP32" names="152,input.19">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="161" name="__module.roberta.encoder.layer.0.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="198192684" size="12288" />
+		<layer id="160" name="self.roberta.encoder.layer.0.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="197926735" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="162" name="__module.roberta.encoder.layer.0.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="161" name="Convert_24448" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="163" name="__module.roberta.encoder.layer.0.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="198204972" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="164" name="__module.roberta.encoder.layer.0.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="198204976" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="165" name="__module.roberta.encoder.layer.0.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="198204972" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="166" name="__module.roberta.encoder.layer.0.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="198204976" size="4" />
+		<layer id="162" name="self.roberta.encoder.layer.0.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="200286031" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="167" name="__module.roberta.encoder.layer.0.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="163" name="Convert_24451" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="168" name="Constant_1059035" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="198204980" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="169" name="Convert_1059036" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="164" name="self.roberta.encoder.layer.0.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="170" name="Constant_1059037" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="200564276" size="3072" />
+		<layer id="165" name="self.roberta.encoder.layer.0.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="200286799" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="171" name="__module.roberta.encoder.layer.0.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="166" name="self.roberta.encoder.layer.0.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="172" name="__module.roberta.encoder.layer.0.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="167" name="self.roberta.encoder.layer.0.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="168" name="__module.roberta.encoder.layer.0.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2375,8 +2463,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="173" name="Constant_8867" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="200567348" size="3072" />
+		<layer id="169" name="Constant_8907" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="200288335" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -2385,7 +2473,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="174" name="__module.roberta.encoder.layer.0.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="170" name="__module.roberta.encoder.layer.0.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2400,14 +2488,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="163,input.5">
+				<port id="2" precision="FP32" names="157,input.21">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="175" name="__module.roberta.encoder.layer.0.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="171" name="__module.roberta.encoder.layer.0.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2422,22 +2510,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="165">
+				<port id="2" precision="FP32" names="159,input.23">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="176" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="172" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="177" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="173" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2457,8 +2545,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="178" name="Constant_8868" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="200570420" size="3072" />
+		<layer id="174" name="Constant_8908" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="200291407" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -2467,7 +2555,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="179" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="175" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2489,8 +2577,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="180" name="Constant_8869" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="200573492" size="3072" />
+		<layer id="176" name="Constant_8909" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="200294479" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -2499,7 +2587,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="181" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="177" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2514,143 +2602,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="169,hidden_states.7">
+				<port id="2" precision="FP32" names="163,hidden_states.7">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="182" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="200576564" size="3072" />
+		<layer id="178" name="self.roberta.encoder.layer.1.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="200297551" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="183" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="179" name="Convert_140100" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="184" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="200579636" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="185" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="200579640" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="186" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="200579636" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="187" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="200579640" size="4" />
+		<layer id="180" name="self.roberta.encoder.layer.1.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="200887375" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="188" name="__module.roberta.encoder.layer.0.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="181" name="Convert_140103" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="189" name="Constant_1059039" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="200579644" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="182" name="self.roberta.encoder.layer.1.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="190" name="Convert_1059040" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="191" name="Constant_1059041" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="201169468" size="3072" />
+		<layer id="183" name="self.roberta.encoder.layer.1.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="200888143" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="192" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="184" name="self.roberta.encoder.layer.1.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="185" name="self.roberta.encoder.layer.1.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="193" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="186" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2671,8 +2740,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="194" name="Constant_8870" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="201172540" size="3072" />
+		<layer id="187" name="Constant_8910" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="200889679" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -2681,7 +2750,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="195" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="188" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2696,67 +2765,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="182,x.13">
+				<port id="2" precision="FP32" names="176">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="196" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="201175612" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="197" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="201175616" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="198" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="201175612" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="199" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="201175616" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="200" name="__module.roberta.encoder.layer.1.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="201" name="__module.roberta.encoder.layer.1.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="189" name="Constant_9261" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="202" name="__module.roberta.encoder.layer.1.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="190" name="__module.roberta.encoder.layer.1.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2769,7 +2793,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="186,x.15">
+				<port id="2" precision="FP32" names="178">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -2777,15 +2801,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="203" name="Constant_507" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="191" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="187">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="204" name="__module.roberta.encoder.layer.1.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="192" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -2793,12 +2817,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="188">
+				<port id="2" precision="FP32" names="179,query_layer.3">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -2806,165 +2830,178 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="205" name="Constant_1059043" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="201175620" size="589824" />
+		<layer id="193" name="self.roberta.encoder.layer.1.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="200892751" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="206" name="Convert_1059044" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="194" name="Convert_136112" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="207" name="Constant_1059045" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="201765444" size="3072" />
+		<layer id="195" name="self.roberta.encoder.layer.1.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="201482575" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="208" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="196" name="Convert_136115" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>768</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="209" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
+		<layer id="197" name="self.roberta.encoder.layer.1.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
 					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="210" name="Constant_8871" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="201768516" size="3072" />
+		<layer id="198" name="self.roberta.encoder.layer.1.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="201483343" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="211" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+		<layer id="199" name="self.roberta.encoder.layer.1.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="191,x.17">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="212" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="201771588" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="213" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="201771592" size="4" />
+		<layer id="200" name="self.roberta.encoder.layer.1.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="214" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="201771588" size="4" />
+		<layer id="201" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
+			<input>
+				<port id="0" precision="FP32">
+					<dim>-1</dim>
+					<dim>-1</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="2" precision="FP32">
+					<dim>-1</dim>
+					<dim>-1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="215" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="201771592" size="4" />
+		<layer id="202" name="Constant_8911" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="201484879" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="216" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="203" name="__module.roberta.encoder.layer.1.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="182">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="217" name="__module.roberta.encoder.layer.1.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="204" name="Constant_9262" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="218" name="__module.roberta.encoder.layer.1.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="205" name="__module.roberta.encoder.layer.1.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -2977,7 +3014,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="195,x.19">
+				<port id="2" precision="FP32" names="184">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -2985,15 +3022,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="219" name="Constant_541" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="206" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="196">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="220" name="__module.roberta.encoder.layer.1.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="207" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -3001,12 +3038,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="197">
+				<port id="2" precision="FP32" names="185,key_layer.3">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -3014,59 +3051,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="221" name="Constant_1059047" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="201771596" size="589824" />
+		<layer id="208" name="self.roberta.encoder.layer.1.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="201487951" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="222" name="Convert_1059048" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="209" name="Convert_132124" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="223" name="Constant_1059049" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="202361420" size="3072" />
+		<layer id="210" name="self.roberta.encoder.layer.1.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="202077775" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="211" name="Convert_132127" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="224" name="__module.roberta.encoder.layer.1.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="212" name="self.roberta.encoder.layer.1.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="213" name="self.roberta.encoder.layer.1.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="202078543" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="214" name="self.roberta.encoder.layer.1.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="215" name="self.roberta.encoder.layer.1.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="225" name="__module.roberta.encoder.layer.1.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="216" name="__module.roberta.encoder.layer.1.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3087,8 +3182,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="226" name="Constant_8872" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="202364492" size="3072" />
+		<layer id="217" name="Constant_8912" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="202080079" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -3097,7 +3192,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="227" name="__module.roberta.encoder.layer.1.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="218" name="__module.roberta.encoder.layer.1.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3112,22 +3207,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="200,x.21">
+				<port id="2" precision="FP32" names="188">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="228" name="__module.roberta.encoder.layer.1.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="219" name="Constant_9263" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="229" name="__module.roberta.encoder.layer.1.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="220" name="__module.roberta.encoder.layer.1.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3140,7 +3235,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="204,x.23">
+				<port id="2" precision="FP32" names="190">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -3148,15 +3243,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="230" name="Constant_575" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="221" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="205">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="231" name="__module.roberta.encoder.layer.1.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="222" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -3164,12 +3259,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="206">
+				<port id="2" precision="FP32" names="191,value_layer.3">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -3177,7 +3272,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="232" name="__module.roberta.encoder.layer.1.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="223" name="__module.roberta.encoder.layer.1.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3206,7 +3301,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="207,attn_output.5">
+				<port id="4" precision="FP32" names="192,attn_output.5">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -3214,15 +3309,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="233" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="224" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="234" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="225" name="__module.roberta.encoder.layer.1.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -3235,7 +3330,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="208,attn_output.7">
+				<port id="2" precision="FP32" names="193,attn_output.7">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -3243,15 +3338,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="235" name="Constant_9170" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="226" name="Constant_9264" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="236" name="__module.roberta.encoder.layer.1.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="227" name="__module.roberta.encoder.layer.1.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3265,143 +3360,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="210">
+				<port id="2" precision="FP32" names="195,input.25">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="237" name="__module.roberta.encoder.layer.1.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="202367564" size="3072" />
+		<layer id="228" name="self.roberta.encoder.layer.1.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="202083151" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="238" name="__module.roberta.encoder.layer.1.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="229" name="Convert_144088" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="239" name="__module.roberta.encoder.layer.1.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="202370636" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="240" name="__module.roberta.encoder.layer.1.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="202370640" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="241" name="__module.roberta.encoder.layer.1.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="202370636" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="242" name="__module.roberta.encoder.layer.1.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="202370640" size="4" />
+		<layer id="230" name="self.roberta.encoder.layer.1.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="202672975" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="243" name="__module.roberta.encoder.layer.1.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="231" name="Convert_144091" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="244" name="Constant_1059051" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="202370644" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="232" name="self.roberta.encoder.layer.1.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="245" name="Convert_1059052" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="246" name="Constant_1059053" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="202960468" size="3072" />
+		<layer id="233" name="self.roberta.encoder.layer.1.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="202673743" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="247" name="__module.roberta.encoder.layer.1.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="234" name="self.roberta.encoder.layer.1.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="235" name="self.roberta.encoder.layer.1.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="248" name="__module.roberta.encoder.layer.1.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="236" name="__module.roberta.encoder.layer.1.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3422,8 +3498,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="249" name="Constant_8873" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="202963540" size="3072" />
+		<layer id="237" name="Constant_8913" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="202675279" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -3432,7 +3508,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="250" name="__module.roberta.encoder.layer.1.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="238" name="__module.roberta.encoder.layer.1.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3447,14 +3523,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="215,input.7">
+				<port id="2" precision="FP32" names="200,input.27">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="251" name="__module.roberta.encoder.layer.1.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="239" name="__module.roberta.encoder.layer.1.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3469,22 +3545,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="217">
+				<port id="2" precision="FP32" names="202,input.29">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="252" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="240" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="253" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="241" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3504,8 +3580,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="254" name="Constant_8874" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="202966612" size="3072" />
+		<layer id="242" name="Constant_8914" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="202678351" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -3514,7 +3590,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="255" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="243" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3536,8 +3612,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="256" name="Constant_8875" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="202969684" size="3072" />
+		<layer id="244" name="Constant_8915" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="202681423" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -3546,7 +3622,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="257" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="245" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3561,143 +3637,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="221,input_tensor.3">
+				<port id="2" precision="FP32" names="206,input.31">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="258" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="202972756" size="3072" />
+		<layer id="246" name="self.roberta.encoder.layer.1.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="202684495" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="259" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="247" name="Convert_28436" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="260" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="202975828" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="261" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="202975832" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="262" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="202975828" size="4" />
+		<layer id="248" name="self.roberta.encoder.layer.1.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="205043791" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="263" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="202975832" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="264" name="__module.roberta.encoder.layer.1.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="249" name="Convert_28439" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="265" name="Constant_1059055" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="202975836" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="266" name="Convert_1059056" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="250" name="self.roberta.encoder.layer.1.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="267" name="Constant_1059057" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="205335132" size="12288" />
+		<layer id="251" name="self.roberta.encoder.layer.1.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="205046863" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="268" name="__module.roberta.encoder.layer.1.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="252" name="self.roberta.encoder.layer.1.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="253" name="self.roberta.encoder.layer.1.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="269" name="__module.roberta.encoder.layer.1.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="254" name="__module.roberta.encoder.layer.1.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3718,8 +3775,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="270" name="Constant_8876" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="205347420" size="12288" />
+		<layer id="255" name="Constant_8916" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="205053007" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -3728,7 +3785,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="271" name="__module.roberta.encoder.layer.1.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="256" name="__module.roberta.encoder.layer.1.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3743,14 +3800,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="225">
+				<port id="2" precision="FP32" names="210,input.33">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="272" name="__module.roberta.encoder.layer.1.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="257" name="__module.roberta.encoder.layer.1.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3760,143 +3817,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="226">
+				<port id="1" precision="FP32" names="211,input.35">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="273" name="__module.roberta.encoder.layer.1.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="205359708" size="12288" />
+		<layer id="258" name="self.roberta.encoder.layer.1.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="205065295" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="274" name="__module.roberta.encoder.layer.1.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="259" name="Convert_32424" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="275" name="__module.roberta.encoder.layer.1.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="205371996" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="276" name="__module.roberta.encoder.layer.1.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="205372000" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="277" name="__module.roberta.encoder.layer.1.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="205371996" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="278" name="__module.roberta.encoder.layer.1.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="205372000" size="4" />
+		<layer id="260" name="self.roberta.encoder.layer.1.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="207424591" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="279" name="__module.roberta.encoder.layer.1.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="261" name="Convert_32427" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="280" name="Constant_1059059" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="205372004" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="281" name="Convert_1059060" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="262" name="self.roberta.encoder.layer.1.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="282" name="Constant_1059061" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="207731300" size="3072" />
+		<layer id="263" name="self.roberta.encoder.layer.1.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="207425359" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="283" name="__module.roberta.encoder.layer.1.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="264" name="self.roberta.encoder.layer.1.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="265" name="self.roberta.encoder.layer.1.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="284" name="__module.roberta.encoder.layer.1.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="266" name="__module.roberta.encoder.layer.1.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3917,8 +3955,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="285" name="Constant_8877" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="207734372" size="3072" />
+		<layer id="267" name="Constant_8917" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="207426895" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -3927,7 +3965,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="286" name="__module.roberta.encoder.layer.1.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="268" name="__module.roberta.encoder.layer.1.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3942,14 +3980,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="231,input.9">
+				<port id="2" precision="FP32" names="216,input.37">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="287" name="__module.roberta.encoder.layer.1.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="269" name="__module.roberta.encoder.layer.1.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3964,22 +4002,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="233">
+				<port id="2" precision="FP32" names="218,input.39">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="288" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="270" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="289" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="271" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -3999,8 +4037,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="290" name="Constant_8878" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="207737444" size="3072" />
+		<layer id="272" name="Constant_8918" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="207429967" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -4009,7 +4047,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="291" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="273" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4031,8 +4069,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="292" name="Constant_8879" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="207740516" size="3072" />
+		<layer id="274" name="Constant_8919" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="207433039" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -4041,7 +4079,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="293" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="275" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4056,143 +4094,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="237,hidden_states.13">
+				<port id="2" precision="FP32" names="222,hidden_states.13">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="294" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="207743588" size="3072" />
+		<layer id="276" name="self.roberta.encoder.layer.2.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="207436111" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="295" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="277" name="Convert_156052" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="296" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="207746660" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="297" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="207746664" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="298" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="207746660" size="4" />
+		<layer id="278" name="self.roberta.encoder.layer.2.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="208025935" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="299" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="207746664" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="300" name="__module.roberta.encoder.layer.1.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="279" name="Convert_156055" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="301" name="Constant_1059063" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="207746668" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="280" name="self.roberta.encoder.layer.2.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="302" name="Convert_1059064" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="303" name="Constant_1059065" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="208336492" size="3072" />
+		<layer id="281" name="self.roberta.encoder.layer.2.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="208026703" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="304" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="282" name="self.roberta.encoder.layer.2.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="283" name="self.roberta.encoder.layer.2.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="305" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="284" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4213,8 +4232,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="306" name="Constant_8880" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="208339564" size="3072" />
+		<layer id="285" name="Constant_8920" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="208028239" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -4223,7 +4242,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="307" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="286" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4238,67 +4257,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="250,x.25">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="308" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="208342636" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="309" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="208342640" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="310" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="208342636" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="311" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="208342640" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="312" name="__module.roberta.encoder.layer.2.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="235">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="313" name="__module.roberta.encoder.layer.2.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="287" name="Constant_9265" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="314" name="__module.roberta.encoder.layer.2.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="288" name="__module.roberta.encoder.layer.2.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4311,7 +4285,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="254,x.27">
+				<port id="2" precision="FP32" names="237">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -4319,15 +4293,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="315" name="Constant_741" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="289" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="255">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="316" name="__module.roberta.encoder.layer.2.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="290" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -4335,12 +4309,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="256">
+				<port id="2" precision="FP32" names="238,query_layer.5">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -4348,91 +4322,118 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="317" name="Constant_1059067" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="208342644" size="589824" />
+		<layer id="291" name="self.roberta.encoder.layer.2.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="208031311" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="318" name="Convert_1059068" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="292" name="Convert_152064" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="319" name="Constant_1059069" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="208932468" size="3072" />
+		<layer id="293" name="self.roberta.encoder.layer.2.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="208621135" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="320" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="294" name="Convert_152067" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="295" name="self.roberta.encoder.layer.2.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="321" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+		<layer id="296" name="self.roberta.encoder.layer.2.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="208621903" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32">
+			</output>
+		</layer>
+		<layer id="297" name="self.roberta.encoder.layer.2.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="322" name="Constant_8881" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="208935540" size="3072" />
+		<layer id="298" name="self.roberta.encoder.layer.2.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="323" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/Add" type="Add" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="299" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -4440,73 +4441,59 @@
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="259,x.29">
+				<port id="2" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="324" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="208938612" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="325" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="208938616" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="326" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="208938612" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="327" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="208938616" size="4" />
+		<layer id="300" name="Constant_8921" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="208623439" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="328" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="301" name="__module.roberta.encoder.layer.2.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="241">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="329" name="__module.roberta.encoder.layer.2.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="302" name="Constant_9266" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="330" name="__module.roberta.encoder.layer.2.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="303" name="__module.roberta.encoder.layer.2.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4519,7 +4506,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="263,x.31">
+				<port id="2" precision="FP32" names="243">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -4527,15 +4514,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="331" name="Constant_775" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="304" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="264">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="332" name="__module.roberta.encoder.layer.2.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="305" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -4543,12 +4530,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="265">
+				<port id="2" precision="FP32" names="244,key_layer.5">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -4556,59 +4543,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="333" name="Constant_1059071" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="208938620" size="589824" />
+		<layer id="306" name="self.roberta.encoder.layer.2.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="208626511" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="334" name="Convert_1059072" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="307" name="Convert_148076" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="335" name="Constant_1059073" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="209528444" size="3072" />
+		<layer id="308" name="self.roberta.encoder.layer.2.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="209216335" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="309" name="Convert_148079" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="336" name="__module.roberta.encoder.layer.2.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="310" name="self.roberta.encoder.layer.2.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="311" name="self.roberta.encoder.layer.2.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="209217103" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="312" name="self.roberta.encoder.layer.2.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="313" name="self.roberta.encoder.layer.2.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="337" name="__module.roberta.encoder.layer.2.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="314" name="__module.roberta.encoder.layer.2.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4629,8 +4674,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="338" name="Constant_8882" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="209531516" size="3072" />
+		<layer id="315" name="Constant_8922" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="209218639" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -4639,7 +4684,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="339" name="__module.roberta.encoder.layer.2.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="316" name="__module.roberta.encoder.layer.2.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4654,22 +4699,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="268,x.33">
+				<port id="2" precision="FP32" names="247">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="340" name="__module.roberta.encoder.layer.2.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="317" name="Constant_9267" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="341" name="__module.roberta.encoder.layer.2.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="318" name="__module.roberta.encoder.layer.2.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4682,7 +4727,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="272,x.35">
+				<port id="2" precision="FP32" names="249">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -4690,15 +4735,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="342" name="Constant_809" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="319" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="273">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="343" name="__module.roberta.encoder.layer.2.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="320" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -4706,12 +4751,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="274">
+				<port id="2" precision="FP32" names="250,value_layer.5">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -4719,7 +4764,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="344" name="__module.roberta.encoder.layer.2.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="321" name="__module.roberta.encoder.layer.2.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4748,7 +4793,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="275,attn_output.9">
+				<port id="4" precision="FP32" names="251,attn_output.9">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -4756,15 +4801,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="345" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="322" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="346" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="323" name="__module.roberta.encoder.layer.2.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -4777,7 +4822,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="276,attn_output.11">
+				<port id="2" precision="FP32" names="252,attn_output.11">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -4785,15 +4830,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="347" name="Constant_9171" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="324" name="Constant_9268" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="348" name="__module.roberta.encoder.layer.2.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="325" name="__module.roberta.encoder.layer.2.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4807,143 +4852,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="278">
+				<port id="2" precision="FP32" names="254,input.41">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="349" name="__module.roberta.encoder.layer.2.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="209534588" size="3072" />
+		<layer id="326" name="self.roberta.encoder.layer.2.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="209221711" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="350" name="__module.roberta.encoder.layer.2.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="327" name="Convert_160040" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="351" name="__module.roberta.encoder.layer.2.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="209537660" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="352" name="__module.roberta.encoder.layer.2.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="209537664" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="353" name="__module.roberta.encoder.layer.2.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="209537660" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="354" name="__module.roberta.encoder.layer.2.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="209537664" size="4" />
+		<layer id="328" name="self.roberta.encoder.layer.2.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="209811535" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="355" name="__module.roberta.encoder.layer.2.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="329" name="Convert_160043" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="356" name="Constant_1059075" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="209537668" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="330" name="self.roberta.encoder.layer.2.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="357" name="Convert_1059076" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="358" name="Constant_1059077" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="210127492" size="3072" />
+		<layer id="331" name="self.roberta.encoder.layer.2.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="209812303" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="359" name="__module.roberta.encoder.layer.2.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="332" name="self.roberta.encoder.layer.2.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="333" name="self.roberta.encoder.layer.2.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="360" name="__module.roberta.encoder.layer.2.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="334" name="__module.roberta.encoder.layer.2.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4964,8 +4990,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="361" name="Constant_8883" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="210130564" size="3072" />
+		<layer id="335" name="Constant_8923" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="209813839" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -4974,7 +5000,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="362" name="__module.roberta.encoder.layer.2.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="336" name="__module.roberta.encoder.layer.2.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -4989,14 +5015,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="283,input.11">
+				<port id="2" precision="FP32" names="259,input.43">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="363" name="__module.roberta.encoder.layer.2.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="337" name="__module.roberta.encoder.layer.2.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5011,22 +5037,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="285">
+				<port id="2" precision="FP32" names="261,input.45">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="364" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="338" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="365" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="339" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5046,8 +5072,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="366" name="Constant_8884" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="210133636" size="3072" />
+		<layer id="340" name="Constant_8924" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="209816911" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -5056,7 +5082,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="367" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="341" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5078,8 +5104,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="368" name="Constant_8885" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="210136708" size="3072" />
+		<layer id="342" name="Constant_8925" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="209819983" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -5088,7 +5114,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="369" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="343" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5103,143 +5129,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="289,input_tensor.5">
+				<port id="2" precision="FP32" names="265,input.47">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="370" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="210139780" size="3072" />
+		<layer id="344" name="self.roberta.encoder.layer.2.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="209823055" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="371" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="345" name="Convert_36412" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="372" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="210142852" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="373" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="210142856" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="374" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="210142852" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="375" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="210142856" size="4" />
+		<layer id="346" name="self.roberta.encoder.layer.2.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="212182351" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="376" name="__module.roberta.encoder.layer.2.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="347" name="Convert_36415" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="377" name="Constant_1059079" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="210142860" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="378" name="Convert_1059080" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="348" name="self.roberta.encoder.layer.2.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="379" name="Constant_1059081" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="212502156" size="12288" />
+		<layer id="349" name="self.roberta.encoder.layer.2.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="212185423" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="380" name="__module.roberta.encoder.layer.2.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="350" name="self.roberta.encoder.layer.2.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="351" name="self.roberta.encoder.layer.2.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="381" name="__module.roberta.encoder.layer.2.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="352" name="__module.roberta.encoder.layer.2.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5260,8 +5267,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="382" name="Constant_8886" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="212514444" size="12288" />
+		<layer id="353" name="Constant_8926" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="212191567" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -5270,7 +5277,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="383" name="__module.roberta.encoder.layer.2.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="354" name="__module.roberta.encoder.layer.2.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5285,14 +5292,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="293">
+				<port id="2" precision="FP32" names="269,input.49">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="384" name="__module.roberta.encoder.layer.2.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="355" name="__module.roberta.encoder.layer.2.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5302,147 +5309,128 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="294">
+				<port id="1" precision="FP32" names="270,input.51">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="385" name="__module.roberta.encoder.layer.2.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="212526732" size="12288" />
+		<layer id="356" name="self.roberta.encoder.layer.2.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="212203855" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="386" name="__module.roberta.encoder.layer.2.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="357" name="Convert_40400" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="387" name="__module.roberta.encoder.layer.2.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="212539020" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="388" name="__module.roberta.encoder.layer.2.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="212539024" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="389" name="__module.roberta.encoder.layer.2.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="212539020" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="390" name="__module.roberta.encoder.layer.2.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="212539024" size="4" />
+		<layer id="358" name="self.roberta.encoder.layer.2.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="214563151" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="391" name="__module.roberta.encoder.layer.2.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="359" name="Convert_40403" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="392" name="Constant_1059083" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="212539028" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="393" name="Convert_1059084" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="360" name="self.roberta.encoder.layer.2.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="394" name="Constant_1059085" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="214898324" size="3072" />
+		<layer id="361" name="self.roberta.encoder.layer.2.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="214563919" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="395" name="__module.roberta.encoder.layer.2.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="362" name="self.roberta.encoder.layer.2.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="396" name="__module.roberta.encoder.layer.2.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
+		<layer id="363" name="self.roberta.encoder.layer.2.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="364" name="__module.roberta.encoder.layer.2.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
+			<input>
+				<port id="0" precision="FP32">
+					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
@@ -5459,8 +5447,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="397" name="Constant_8887" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="214901396" size="3072" />
+		<layer id="365" name="Constant_8927" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="214565455" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -5469,7 +5457,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="398" name="__module.roberta.encoder.layer.2.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="366" name="__module.roberta.encoder.layer.2.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5484,14 +5472,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="299,input.13">
+				<port id="2" precision="FP32" names="275,input.53">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="399" name="__module.roberta.encoder.layer.2.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="367" name="__module.roberta.encoder.layer.2.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5506,22 +5494,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="301">
+				<port id="2" precision="FP32" names="277,input.55">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="400" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="368" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="401" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="369" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5541,8 +5529,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="402" name="Constant_8888" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="214904468" size="3072" />
+		<layer id="370" name="Constant_8928" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="214568527" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -5551,7 +5539,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="403" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="371" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5573,8 +5561,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="404" name="Constant_8889" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="214907540" size="3072" />
+		<layer id="372" name="Constant_8929" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="214571599" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -5583,7 +5571,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="405" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="373" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5598,143 +5586,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="305,hidden_states.19">
+				<port id="2" precision="FP32" names="281,hidden_states.19">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="406" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="214910612" size="3072" />
+		<layer id="374" name="self.roberta.encoder.layer.3.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="214574671" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="407" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="375" name="Convert_172004" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="408" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="214913684" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="409" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="214913688" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="410" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="214913684" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="411" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="214913688" size="4" />
+		<layer id="376" name="self.roberta.encoder.layer.3.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="215164495" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="412" name="__module.roberta.encoder.layer.2.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="377" name="Convert_172007" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="413" name="Constant_1059087" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="214913692" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="378" name="self.roberta.encoder.layer.3.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="414" name="Convert_1059088" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="415" name="Constant_1059089" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="215503516" size="3072" />
+		<layer id="379" name="self.roberta.encoder.layer.3.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="215165263" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="416" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="380" name="self.roberta.encoder.layer.3.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="381" name="self.roberta.encoder.layer.3.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="417" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="382" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5755,8 +5724,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="418" name="Constant_8890" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="215506588" size="3072" />
+		<layer id="383" name="Constant_8930" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="215166799" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -5765,7 +5734,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="419" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="384" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5780,67 +5749,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="318,x.37">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="420" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="215509660" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="421" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="215509664" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="422" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="215509660" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="423" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="215509664" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="424" name="__module.roberta.encoder.layer.3.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="294">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="425" name="__module.roberta.encoder.layer.3.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="385" name="Constant_9269" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="426" name="__module.roberta.encoder.layer.3.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="386" name="__module.roberta.encoder.layer.3.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -5853,7 +5777,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="322,x.39">
+				<port id="2" precision="FP32" names="296">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -5861,15 +5785,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="427" name="Constant_975" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="387" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="323">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="428" name="__module.roberta.encoder.layer.3.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="388" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -5877,12 +5801,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="324">
+				<port id="2" precision="FP32" names="297,query_layer.7">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -5890,91 +5814,118 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="429" name="Constant_1059091" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="215509668" size="589824" />
+		<layer id="389" name="self.roberta.encoder.layer.3.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="215169871" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="430" name="Convert_1059092" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="390" name="Convert_168016" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="431" name="Constant_1059093" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="216099492" size="3072" />
+		<layer id="391" name="self.roberta.encoder.layer.3.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="215759695" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="392" name="Convert_168019" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="432" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="393" name="self.roberta.encoder.layer.3.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="433" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+		<layer id="394" name="self.roberta.encoder.layer.3.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="215760463" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32">
+			</output>
+		</layer>
+		<layer id="395" name="self.roberta.encoder.layer.3.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="434" name="Constant_8891" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="216102564" size="3072" />
+		<layer id="396" name="self.roberta.encoder.layer.3.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="435" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/Add" type="Add" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="397" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -5982,73 +5933,59 @@
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="327,x.41">
+				<port id="2" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="436" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="216105636" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="437" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="216105640" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="438" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="216105636" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="439" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="216105640" size="4" />
+		<layer id="398" name="Constant_8931" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="215761999" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="440" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="399" name="__module.roberta.encoder.layer.3.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="300">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="441" name="__module.roberta.encoder.layer.3.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="400" name="Constant_9270" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="442" name="__module.roberta.encoder.layer.3.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="401" name="__module.roberta.encoder.layer.3.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6061,7 +5998,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="331,x.43">
+				<port id="2" precision="FP32" names="302">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -6069,15 +6006,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="443" name="Constant_1009" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="402" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="332">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="444" name="__module.roberta.encoder.layer.3.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="403" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -6085,12 +6022,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="333">
+				<port id="2" precision="FP32" names="303,key_layer.7">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -6098,59 +6035,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="445" name="Constant_1059095" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="216105644" size="589824" />
+		<layer id="404" name="self.roberta.encoder.layer.3.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="215765071" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="446" name="Convert_1059096" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="405" name="Convert_164028" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="447" name="Constant_1059097" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="216695468" size="3072" />
+		<layer id="406" name="self.roberta.encoder.layer.3.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="216354895" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="407" name="Convert_164031" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="448" name="__module.roberta.encoder.layer.3.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="408" name="self.roberta.encoder.layer.3.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="409" name="self.roberta.encoder.layer.3.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="216355663" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="410" name="self.roberta.encoder.layer.3.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="411" name="self.roberta.encoder.layer.3.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="449" name="__module.roberta.encoder.layer.3.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="412" name="__module.roberta.encoder.layer.3.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6171,8 +6166,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="450" name="Constant_8892" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="216698540" size="3072" />
+		<layer id="413" name="Constant_8932" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="216357199" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -6181,7 +6176,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="451" name="__module.roberta.encoder.layer.3.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="414" name="__module.roberta.encoder.layer.3.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6196,22 +6191,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="336,x.45">
+				<port id="2" precision="FP32" names="306">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="452" name="__module.roberta.encoder.layer.3.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="415" name="Constant_9271" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="453" name="__module.roberta.encoder.layer.3.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="416" name="__module.roberta.encoder.layer.3.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6224,7 +6219,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="340,x.47">
+				<port id="2" precision="FP32" names="308">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -6232,15 +6227,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="454" name="Constant_1043" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="417" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="341">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="455" name="__module.roberta.encoder.layer.3.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="418" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -6248,12 +6243,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="342">
+				<port id="2" precision="FP32" names="309,value_layer.7">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -6261,7 +6256,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="456" name="__module.roberta.encoder.layer.3.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="419" name="__module.roberta.encoder.layer.3.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6290,7 +6285,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="343,attn_output.13">
+				<port id="4" precision="FP32" names="310,attn_output.13">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -6298,15 +6293,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="457" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="420" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="458" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="421" name="__module.roberta.encoder.layer.3.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -6319,7 +6314,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="344,attn_output.15">
+				<port id="2" precision="FP32" names="311,attn_output.15">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -6327,15 +6322,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="459" name="Constant_9172" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="422" name="Constant_9272" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="460" name="__module.roberta.encoder.layer.3.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="423" name="__module.roberta.encoder.layer.3.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6349,143 +6344,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="346">
+				<port id="2" precision="FP32" names="313,input.57">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="461" name="__module.roberta.encoder.layer.3.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="216701612" size="3072" />
+		<layer id="424" name="self.roberta.encoder.layer.3.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="216360271" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="462" name="__module.roberta.encoder.layer.3.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="425" name="Convert_175992" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="463" name="__module.roberta.encoder.layer.3.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="216704684" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="464" name="__module.roberta.encoder.layer.3.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="216704688" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="465" name="__module.roberta.encoder.layer.3.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="216704684" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="466" name="__module.roberta.encoder.layer.3.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="216704688" size="4" />
+		<layer id="426" name="self.roberta.encoder.layer.3.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="216950095" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="467" name="__module.roberta.encoder.layer.3.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="427" name="Convert_175995" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="468" name="Constant_1059099" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="216704692" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="428" name="self.roberta.encoder.layer.3.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="469" name="Convert_1059100" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="470" name="Constant_1059101" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="217294516" size="3072" />
+		<layer id="429" name="self.roberta.encoder.layer.3.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="216950863" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="471" name="__module.roberta.encoder.layer.3.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="430" name="self.roberta.encoder.layer.3.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="431" name="self.roberta.encoder.layer.3.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="472" name="__module.roberta.encoder.layer.3.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="432" name="__module.roberta.encoder.layer.3.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6506,8 +6482,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="473" name="Constant_8893" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="217297588" size="3072" />
+		<layer id="433" name="Constant_8933" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="216952399" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -6516,7 +6492,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="474" name="__module.roberta.encoder.layer.3.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="434" name="__module.roberta.encoder.layer.3.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6531,14 +6507,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="351,input.15">
+				<port id="2" precision="FP32" names="318,input.59">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="475" name="__module.roberta.encoder.layer.3.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="435" name="__module.roberta.encoder.layer.3.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6553,22 +6529,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="353">
+				<port id="2" precision="FP32" names="320,input.61">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="476" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="436" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="477" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="437" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6588,8 +6564,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="478" name="Constant_8894" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="217300660" size="3072" />
+		<layer id="438" name="Constant_8934" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="216955471" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -6598,7 +6574,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="479" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="439" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6620,8 +6596,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="480" name="Constant_8895" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="217303732" size="3072" />
+		<layer id="440" name="Constant_8935" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="216958543" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -6630,7 +6606,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="481" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="441" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6645,143 +6621,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="357,input_tensor.7">
+				<port id="2" precision="FP32" names="324,input.63">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="482" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="217306804" size="3072" />
+		<layer id="442" name="self.roberta.encoder.layer.3.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="216961615" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="483" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="443" name="Convert_44388" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="484" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="217309876" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="485" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="217309880" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="486" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="217309876" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="487" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="217309880" size="4" />
+		<layer id="444" name="self.roberta.encoder.layer.3.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="219320911" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="488" name="__module.roberta.encoder.layer.3.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="445" name="Convert_44391" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="489" name="Constant_1059103" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="217309884" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="490" name="Convert_1059104" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="446" name="self.roberta.encoder.layer.3.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="491" name="Constant_1059105" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="219669180" size="12288" />
+		<layer id="447" name="self.roberta.encoder.layer.3.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="219323983" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="492" name="__module.roberta.encoder.layer.3.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="448" name="self.roberta.encoder.layer.3.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="449" name="self.roberta.encoder.layer.3.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="493" name="__module.roberta.encoder.layer.3.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="450" name="__module.roberta.encoder.layer.3.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6802,8 +6759,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="494" name="Constant_8896" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="219681468" size="12288" />
+		<layer id="451" name="Constant_8936" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="219330127" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -6812,7 +6769,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="495" name="__module.roberta.encoder.layer.3.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="452" name="__module.roberta.encoder.layer.3.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6827,14 +6784,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="361">
+				<port id="2" precision="FP32" names="328,input.65">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="496" name="__module.roberta.encoder.layer.3.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="453" name="__module.roberta.encoder.layer.3.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -6844,143 +6801,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="362">
+				<port id="1" precision="FP32" names="329,input.67">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="497" name="__module.roberta.encoder.layer.3.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="219693756" size="12288" />
+		<layer id="454" name="self.roberta.encoder.layer.3.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="219342415" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="498" name="__module.roberta.encoder.layer.3.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="455" name="Convert_48376" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="499" name="__module.roberta.encoder.layer.3.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="219706044" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="500" name="__module.roberta.encoder.layer.3.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="219706048" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="501" name="__module.roberta.encoder.layer.3.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="219706044" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="502" name="__module.roberta.encoder.layer.3.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="219706048" size="4" />
+		<layer id="456" name="self.roberta.encoder.layer.3.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="221701711" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="503" name="__module.roberta.encoder.layer.3.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="457" name="Convert_48379" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="504" name="Constant_1059107" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="219706052" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="505" name="Convert_1059108" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="458" name="self.roberta.encoder.layer.3.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="506" name="Constant_1059109" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="222065348" size="3072" />
+		<layer id="459" name="self.roberta.encoder.layer.3.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="221702479" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="507" name="__module.roberta.encoder.layer.3.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="460" name="self.roberta.encoder.layer.3.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="461" name="self.roberta.encoder.layer.3.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="508" name="__module.roberta.encoder.layer.3.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="462" name="__module.roberta.encoder.layer.3.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7001,8 +6939,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="509" name="Constant_8897" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="222068420" size="3072" />
+		<layer id="463" name="Constant_8937" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="221704015" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -7011,7 +6949,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="510" name="__module.roberta.encoder.layer.3.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="464" name="__module.roberta.encoder.layer.3.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7026,14 +6964,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="367,input.17">
+				<port id="2" precision="FP32" names="334,input.69">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="511" name="__module.roberta.encoder.layer.3.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="465" name="__module.roberta.encoder.layer.3.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7048,22 +6986,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="369">
+				<port id="2" precision="FP32" names="336,input.71">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="512" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="466" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="513" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="467" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7083,8 +7021,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="514" name="Constant_8898" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="222071492" size="3072" />
+		<layer id="468" name="Constant_8938" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="221707087" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -7093,7 +7031,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="515" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="469" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7115,8 +7053,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="516" name="Constant_8899" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="222074564" size="3072" />
+		<layer id="470" name="Constant_8939" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="221710159" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -7125,7 +7063,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="517" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="471" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7140,143 +7078,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="373,hidden_states.25">
+				<port id="2" precision="FP32" names="340,hidden_states.25">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="518" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="222077636" size="3072" />
+		<layer id="472" name="self.roberta.encoder.layer.4.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="221713231" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="519" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="473" name="Convert_187956" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="520" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="222080708" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="521" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="222080712" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="522" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="222080708" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="523" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="222080712" size="4" />
+		<layer id="474" name="self.roberta.encoder.layer.4.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="222303055" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="524" name="__module.roberta.encoder.layer.3.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="475" name="Convert_187959" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="525" name="Constant_1059111" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="222080716" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="476" name="self.roberta.encoder.layer.4.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="526" name="Convert_1059112" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="527" name="Constant_1059113" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="222670540" size="3072" />
+		<layer id="477" name="self.roberta.encoder.layer.4.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="222303823" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="528" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="478" name="self.roberta.encoder.layer.4.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="479" name="self.roberta.encoder.layer.4.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="529" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="480" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7297,8 +7216,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="530" name="Constant_8900" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="222673612" size="3072" />
+		<layer id="481" name="Constant_8940" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="222305359" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -7307,7 +7226,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="531" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="482" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7322,67 +7241,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="386,x.49">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="532" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="222676684" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="533" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="222676688" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="534" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="222676684" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="535" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="222676688" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="536" name="__module.roberta.encoder.layer.4.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="353">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="537" name="__module.roberta.encoder.layer.4.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="483" name="Constant_9273" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="538" name="__module.roberta.encoder.layer.4.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="484" name="__module.roberta.encoder.layer.4.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7395,7 +7269,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="390,x.51">
+				<port id="2" precision="FP32" names="355">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -7403,15 +7277,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="539" name="Constant_1209" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="485" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="391">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="540" name="__module.roberta.encoder.layer.4.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="486" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -7419,12 +7293,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="392">
+				<port id="2" precision="FP32" names="356,query_layer.9">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -7432,91 +7306,118 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="541" name="Constant_1059115" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="222676692" size="589824" />
+		<layer id="487" name="self.roberta.encoder.layer.4.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="222308431" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="542" name="Convert_1059116" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="488" name="Convert_183968" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="543" name="Constant_1059117" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="223266516" size="3072" />
+		<layer id="489" name="self.roberta.encoder.layer.4.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="222898255" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="490" name="Convert_183971" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="544" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="491" name="self.roberta.encoder.layer.4.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="545" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+		<layer id="492" name="self.roberta.encoder.layer.4.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="222899023" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32">
+			</output>
+		</layer>
+		<layer id="493" name="self.roberta.encoder.layer.4.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="546" name="Constant_8901" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="223269588" size="3072" />
+		<layer id="494" name="self.roberta.encoder.layer.4.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="547" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/Add" type="Add" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="495" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -7524,73 +7425,59 @@
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="395,x.53">
+				<port id="2" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="548" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="223272660" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="549" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="223272664" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="550" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="223272660" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="551" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="223272664" size="4" />
+		<layer id="496" name="Constant_8941" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="222900559" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="552" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="497" name="__module.roberta.encoder.layer.4.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="359">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="553" name="__module.roberta.encoder.layer.4.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="498" name="Constant_9274" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="554" name="__module.roberta.encoder.layer.4.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="499" name="__module.roberta.encoder.layer.4.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7603,7 +7490,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="399,x.55">
+				<port id="2" precision="FP32" names="361">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -7611,15 +7498,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="555" name="Constant_1243" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="500" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="400">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="556" name="__module.roberta.encoder.layer.4.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="501" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -7627,12 +7514,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="401">
+				<port id="2" precision="FP32" names="362,key_layer.9">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -7640,59 +7527,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="557" name="Constant_1059119" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="223272668" size="589824" />
+		<layer id="502" name="self.roberta.encoder.layer.4.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="222903631" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="558" name="Convert_1059120" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="503" name="Convert_179980" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="559" name="Constant_1059121" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="223862492" size="3072" />
+		<layer id="504" name="self.roberta.encoder.layer.4.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="223493455" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="505" name="Convert_179983" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="560" name="__module.roberta.encoder.layer.4.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="506" name="self.roberta.encoder.layer.4.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="507" name="self.roberta.encoder.layer.4.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="223494223" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="508" name="self.roberta.encoder.layer.4.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="561" name="__module.roberta.encoder.layer.4.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="509" name="self.roberta.encoder.layer.4.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="510" name="__module.roberta.encoder.layer.4.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7713,8 +7658,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="562" name="Constant_8902" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="223865564" size="3072" />
+		<layer id="511" name="Constant_8942" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="223495759" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -7723,7 +7668,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="563" name="__module.roberta.encoder.layer.4.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="512" name="__module.roberta.encoder.layer.4.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7738,22 +7683,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="404,x.57">
+				<port id="2" precision="FP32" names="365">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="564" name="__module.roberta.encoder.layer.4.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="513" name="Constant_9275" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="565" name="__module.roberta.encoder.layer.4.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="514" name="__module.roberta.encoder.layer.4.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7766,7 +7711,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="408,x.59">
+				<port id="2" precision="FP32" names="367">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -7774,15 +7719,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="566" name="Constant_1277" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="515" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="409">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="567" name="__module.roberta.encoder.layer.4.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="516" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -7790,12 +7735,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="410">
+				<port id="2" precision="FP32" names="368,value_layer.9">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -7803,7 +7748,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="568" name="__module.roberta.encoder.layer.4.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="517" name="__module.roberta.encoder.layer.4.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7832,7 +7777,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="411,attn_output.17">
+				<port id="4" precision="FP32" names="369,attn_output.17">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -7840,15 +7785,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="569" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="518" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="570" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="519" name="__module.roberta.encoder.layer.4.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -7861,7 +7806,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="412,attn_output.19">
+				<port id="2" precision="FP32" names="370,attn_output.19">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -7869,15 +7814,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="571" name="Constant_9173" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="520" name="Constant_9276" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="572" name="__module.roberta.encoder.layer.4.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="521" name="__module.roberta.encoder.layer.4.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -7891,155 +7836,136 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="414">
+				<port id="2" precision="FP32" names="372,input.73">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="573" name="__module.roberta.encoder.layer.4.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="223868636" size="3072" />
+		<layer id="522" name="self.roberta.encoder.layer.4.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="223498831" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="574" name="__module.roberta.encoder.layer.4.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="523" name="Convert_191944" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="575" name="__module.roberta.encoder.layer.4.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="223871708" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="576" name="__module.roberta.encoder.layer.4.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="223871712" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="577" name="__module.roberta.encoder.layer.4.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="223871708" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="578" name="__module.roberta.encoder.layer.4.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="223871712" size="4" />
+		<layer id="524" name="self.roberta.encoder.layer.4.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="224088655" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="579" name="__module.roberta.encoder.layer.4.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="525" name="Convert_191947" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="580" name="Constant_1059123" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="223871716" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="526" name="self.roberta.encoder.layer.4.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="581" name="Convert_1059124" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="582" name="Constant_1059125" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="224461540" size="3072" />
+		<layer id="527" name="self.roberta.encoder.layer.4.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="224089423" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="583" name="__module.roberta.encoder.layer.4.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="528" name="self.roberta.encoder.layer.4.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="584" name="__module.roberta.encoder.layer.4.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
+		<layer id="529" name="self.roberta.encoder.layer.4.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
+			</input>
+			<output>
 				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</input>
+			</output>
+		</layer>
+		<layer id="530" name="__module.roberta.encoder.layer.4.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
+			<input>
+				<port id="0" precision="FP32">
+					<dim>-1</dim>
+					<dim>-1</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
 				<port id="2" precision="FP32">
 					<dim>-1</dim>
@@ -8048,8 +7974,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="585" name="Constant_8903" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="224464612" size="3072" />
+		<layer id="531" name="Constant_8943" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="224090959" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -8058,7 +7984,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="586" name="__module.roberta.encoder.layer.4.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="532" name="__module.roberta.encoder.layer.4.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8073,14 +7999,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="419,input.19">
+				<port id="2" precision="FP32" names="377,input.75">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="587" name="__module.roberta.encoder.layer.4.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="533" name="__module.roberta.encoder.layer.4.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8095,22 +8021,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="421">
+				<port id="2" precision="FP32" names="379,input.77">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="588" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="534" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="589" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="535" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8130,8 +8056,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="590" name="Constant_8904" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="224467684" size="3072" />
+		<layer id="536" name="Constant_8944" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="224094031" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -8140,7 +8066,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="591" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="537" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8162,8 +8088,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="592" name="Constant_8905" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="224470756" size="3072" />
+		<layer id="538" name="Constant_8945" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="224097103" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -8172,7 +8098,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="593" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="539" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8187,143 +8113,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="425,input_tensor.9">
+				<port id="2" precision="FP32" names="383,input.79">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="594" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="224473828" size="3072" />
+		<layer id="540" name="self.roberta.encoder.layer.4.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="224100175" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="595" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="541" name="Convert_52364" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="596" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="224476900" size="4" />
+		<layer id="542" name="self.roberta.encoder.layer.4.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="226459471" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="597" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="224476904" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="598" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="224476900" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="599" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="224476904" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="600" name="__module.roberta.encoder.layer.4.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="543" name="Convert_52367" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="601" name="Constant_1059127" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="224476908" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="602" name="Convert_1059128" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="544" name="self.roberta.encoder.layer.4.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="603" name="Constant_1059129" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="226836204" size="12288" />
+		<layer id="545" name="self.roberta.encoder.layer.4.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="226462543" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="604" name="__module.roberta.encoder.layer.4.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="546" name="self.roberta.encoder.layer.4.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="547" name="self.roberta.encoder.layer.4.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="605" name="__module.roberta.encoder.layer.4.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="548" name="__module.roberta.encoder.layer.4.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8344,8 +8251,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="606" name="Constant_8906" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="226848492" size="12288" />
+		<layer id="549" name="Constant_8946" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="226468687" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -8354,7 +8261,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="607" name="__module.roberta.encoder.layer.4.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="550" name="__module.roberta.encoder.layer.4.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8369,14 +8276,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="429">
+				<port id="2" precision="FP32" names="387,input.81">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="608" name="__module.roberta.encoder.layer.4.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="551" name="__module.roberta.encoder.layer.4.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8386,143 +8293,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="430">
+				<port id="1" precision="FP32" names="388,input.83">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="609" name="__module.roberta.encoder.layer.4.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="226860780" size="12288" />
+		<layer id="552" name="self.roberta.encoder.layer.4.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="226480975" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="610" name="__module.roberta.encoder.layer.4.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="553" name="Convert_56352" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="611" name="__module.roberta.encoder.layer.4.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="226873068" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="612" name="__module.roberta.encoder.layer.4.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="226873072" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="613" name="__module.roberta.encoder.layer.4.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="226873068" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="614" name="__module.roberta.encoder.layer.4.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="226873072" size="4" />
+		<layer id="554" name="self.roberta.encoder.layer.4.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="228840271" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="615" name="__module.roberta.encoder.layer.4.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="555" name="Convert_56355" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="616" name="Constant_1059131" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="226873076" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="617" name="Convert_1059132" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="556" name="self.roberta.encoder.layer.4.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="618" name="Constant_1059133" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="229232372" size="3072" />
+		<layer id="557" name="self.roberta.encoder.layer.4.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="228841039" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="619" name="__module.roberta.encoder.layer.4.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="558" name="self.roberta.encoder.layer.4.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="559" name="self.roberta.encoder.layer.4.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="620" name="__module.roberta.encoder.layer.4.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="560" name="__module.roberta.encoder.layer.4.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8543,8 +8431,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="621" name="Constant_8907" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="229235444" size="3072" />
+		<layer id="561" name="Constant_8947" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="228842575" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -8553,7 +8441,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="622" name="__module.roberta.encoder.layer.4.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="562" name="__module.roberta.encoder.layer.4.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8568,14 +8456,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="435,input.21">
+				<port id="2" precision="FP32" names="393,input.85">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="623" name="__module.roberta.encoder.layer.4.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="563" name="__module.roberta.encoder.layer.4.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8590,22 +8478,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="437">
+				<port id="2" precision="FP32" names="395,input.87">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="624" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="564" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="625" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="565" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8625,8 +8513,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="626" name="Constant_8908" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="229238516" size="3072" />
+		<layer id="566" name="Constant_8948" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="228845647" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -8635,7 +8523,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="627" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="567" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8657,8 +8545,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="628" name="Constant_8909" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="229241588" size="3072" />
+		<layer id="568" name="Constant_8949" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="228848719" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -8667,7 +8555,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="629" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="569" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8682,143 +8570,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="441,hidden_states.31">
+				<port id="2" precision="FP32" names="399,hidden_states.31">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="630" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="229244660" size="3072" />
+		<layer id="570" name="self.roberta.encoder.layer.5.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="228851791" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="631" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="571" name="Convert_203908" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="632" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="229247732" size="4" />
+		<layer id="572" name="self.roberta.encoder.layer.5.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="229441615" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="633" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="229247736" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="634" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="229247732" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="635" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="229247736" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="636" name="__module.roberta.encoder.layer.4.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="573" name="Convert_203911" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="637" name="Constant_1059135" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="229247740" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="574" name="self.roberta.encoder.layer.5.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="638" name="Convert_1059136" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="639" name="Constant_1059137" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="229837564" size="3072" />
+		<layer id="575" name="self.roberta.encoder.layer.5.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="229442383" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="640" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="576" name="self.roberta.encoder.layer.5.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="577" name="self.roberta.encoder.layer.5.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="641" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="578" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8839,8 +8708,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="642" name="Constant_8910" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="229840636" size="3072" />
+		<layer id="579" name="Constant_8950" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="229443919" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -8849,7 +8718,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="643" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="580" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -8864,169 +8733,182 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="454,x.61">
+				<port id="2" precision="FP32" names="412">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="644" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="229843708" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="645" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="229843712" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="646" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="229843708" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="647" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="229843712" size="4" />
+		<layer id="581" name="Constant_9277" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="I64">
+					<dim>4</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="648" name="__module.roberta.encoder.layer.5.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="582" name="__module.roberta.encoder.layer.5.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="I64">
+					<dim>4</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="414">
 					<dim>-1</dim>
 					<dim>-1</dim>
-					<dim>768</dim>
+					<dim>12</dim>
+					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="649" name="__module.roberta.encoder.layer.5.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="583" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="650" name="__module.roberta.encoder.layer.5.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
-			<data special_zero="true" />
+		<layer id="584" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
-					<dim>768</dim>
+					<dim>12</dim>
+					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="458,x.63">
-					<dim>-1</dim>
+				<port id="2" precision="FP32" names="415,query_layer.11">
 					<dim>-1</dim>
 					<dim>12</dim>
+					<dim>-1</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="651" name="Constant_1443" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="585" name="self.roberta.encoder.layer.5.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="229446991" size="589824" />
 			<output>
-				<port id="0" precision="I64" names="459">
-					<dim>4</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="652" name="__module.roberta.encoder.layer.5.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="586" name="Convert_199920" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>12</dim>
-					<dim>64</dim>
-				</port>
-				<port id="1" precision="I64">
-					<dim>4</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="460">
-					<dim>-1</dim>
-					<dim>12</dim>
-					<dim>-1</dim>
-					<dim>64</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="653" name="Constant_1059139" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="229843716" size="589824" />
+		<layer id="587" name="self.roberta.encoder.layer.5.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="230036815" size="768" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="588" name="Convert_199923" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="654" name="Convert_1059140" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="589" name="self.roberta.encoder.layer.5.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="655" name="Constant_1059141" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="230433540" size="3072" />
+		<layer id="590" name="self.roberta.encoder.layer.5.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="230037583" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="656" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="591" name="self.roberta.encoder.layer.5.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="592" name="self.roberta.encoder.layer.5.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="657" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="593" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9047,8 +8929,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="658" name="Constant_8911" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="230436612" size="3072" />
+		<layer id="594" name="Constant_8951" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="230039119" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -9057,7 +8939,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="659" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+		<layer id="595" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9072,67 +8954,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="463,x.65">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="660" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="230439684" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="661" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="230439688" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="662" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="230439684" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="663" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="230439688" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="664" name="__module.roberta.encoder.layer.5.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="418">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="665" name="__module.roberta.encoder.layer.5.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="596" name="Constant_9278" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="666" name="__module.roberta.encoder.layer.5.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="597" name="__module.roberta.encoder.layer.5.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9145,7 +8982,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="467,x.67">
+				<port id="2" precision="FP32" names="420">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -9153,15 +8990,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="667" name="Constant_1477" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="598" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="468">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="668" name="__module.roberta.encoder.layer.5.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="599" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -9169,12 +9006,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="469">
+				<port id="2" precision="FP32" names="421,key_layer.11">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -9182,59 +9019,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="669" name="Constant_1059143" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="230439692" size="589824" />
+		<layer id="600" name="self.roberta.encoder.layer.5.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="230042191" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="670" name="Convert_1059144" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="601" name="Convert_195932" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="671" name="Constant_1059145" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="231029516" size="3072" />
+		<layer id="602" name="self.roberta.encoder.layer.5.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="230632015" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="672" name="__module.roberta.encoder.layer.5.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="603" name="Convert_195935" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="604" name="self.roberta.encoder.layer.5.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="605" name="self.roberta.encoder.layer.5.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="230632783" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="606" name="self.roberta.encoder.layer.5.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="607" name="self.roberta.encoder.layer.5.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="673" name="__module.roberta.encoder.layer.5.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="608" name="__module.roberta.encoder.layer.5.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9255,8 +9150,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="674" name="Constant_8912" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="231032588" size="3072" />
+		<layer id="609" name="Constant_8952" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="230634319" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -9265,7 +9160,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="675" name="__module.roberta.encoder.layer.5.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="610" name="__module.roberta.encoder.layer.5.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9280,22 +9175,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="472,x.69">
+				<port id="2" precision="FP32" names="424">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="676" name="__module.roberta.encoder.layer.5.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="611" name="Constant_9279" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="677" name="__module.roberta.encoder.layer.5.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="612" name="__module.roberta.encoder.layer.5.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9308,7 +9203,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="476,x.71">
+				<port id="2" precision="FP32" names="426">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -9316,15 +9211,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="678" name="Constant_1511" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="613" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="477">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="679" name="__module.roberta.encoder.layer.5.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="614" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -9332,12 +9227,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="478">
+				<port id="2" precision="FP32" names="427,value_layer.11">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -9345,7 +9240,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="680" name="__module.roberta.encoder.layer.5.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="615" name="__module.roberta.encoder.layer.5.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9374,7 +9269,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="479,attn_output.21">
+				<port id="4" precision="FP32" names="428,attn_output.21">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -9382,15 +9277,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="681" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="616" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="682" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="617" name="__module.roberta.encoder.layer.5.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -9403,7 +9298,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="480,attn_output.23">
+				<port id="2" precision="FP32" names="429,attn_output.23">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -9411,15 +9306,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="683" name="Constant_9174" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="618" name="Constant_9280" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="684" name="__module.roberta.encoder.layer.5.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="619" name="__module.roberta.encoder.layer.5.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9433,143 +9328,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="482">
+				<port id="2" precision="FP32" names="431,input.89">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="685" name="__module.roberta.encoder.layer.5.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="231035660" size="3072" />
+		<layer id="620" name="self.roberta.encoder.layer.5.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="230637391" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="686" name="__module.roberta.encoder.layer.5.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="621" name="Convert_207896" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="687" name="__module.roberta.encoder.layer.5.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="231038732" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="688" name="__module.roberta.encoder.layer.5.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="231038736" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="689" name="__module.roberta.encoder.layer.5.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="231038732" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="690" name="__module.roberta.encoder.layer.5.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="231038736" size="4" />
+		<layer id="622" name="self.roberta.encoder.layer.5.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="231227215" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="691" name="__module.roberta.encoder.layer.5.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="623" name="Convert_207899" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="692" name="Constant_1059147" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="231038740" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="624" name="self.roberta.encoder.layer.5.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="693" name="Convert_1059148" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="694" name="Constant_1059149" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="231628564" size="3072" />
+		<layer id="625" name="self.roberta.encoder.layer.5.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="231227983" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="695" name="__module.roberta.encoder.layer.5.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="626" name="self.roberta.encoder.layer.5.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="627" name="self.roberta.encoder.layer.5.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="696" name="__module.roberta.encoder.layer.5.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="628" name="__module.roberta.encoder.layer.5.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9590,8 +9466,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="697" name="Constant_8913" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="231631636" size="3072" />
+		<layer id="629" name="Constant_8953" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="231229519" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -9600,7 +9476,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="698" name="__module.roberta.encoder.layer.5.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="630" name="__module.roberta.encoder.layer.5.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9615,14 +9491,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="487,input.23">
+				<port id="2" precision="FP32" names="436,input.91">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="699" name="__module.roberta.encoder.layer.5.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="631" name="__module.roberta.encoder.layer.5.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9637,22 +9513,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="489">
+				<port id="2" precision="FP32" names="438,input.93">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="700" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="632" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="701" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="633" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9672,8 +9548,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="702" name="Constant_8914" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="231634708" size="3072" />
+		<layer id="634" name="Constant_8954" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="231232591" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -9682,7 +9558,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="703" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="635" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9704,8 +9580,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="704" name="Constant_8915" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="231637780" size="3072" />
+		<layer id="636" name="Constant_8955" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="231235663" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -9714,7 +9590,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="705" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="637" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9729,143 +9605,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="493,input_tensor.11">
+				<port id="2" precision="FP32" names="442,input.95">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="706" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="231640852" size="3072" />
+		<layer id="638" name="self.roberta.encoder.layer.5.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="231238735" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="707" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="639" name="Convert_60340" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="708" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="231643924" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="709" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="231643928" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="710" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="231643924" size="4" />
+		<layer id="640" name="self.roberta.encoder.layer.5.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="233598031" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="711" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="231643928" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="712" name="__module.roberta.encoder.layer.5.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="641" name="Convert_60343" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="713" name="Constant_1059151" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="231643932" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="714" name="Convert_1059152" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="642" name="self.roberta.encoder.layer.5.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="715" name="Constant_1059153" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="234003228" size="12288" />
+		<layer id="643" name="self.roberta.encoder.layer.5.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="233601103" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="716" name="__module.roberta.encoder.layer.5.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="644" name="self.roberta.encoder.layer.5.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="645" name="self.roberta.encoder.layer.5.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="717" name="__module.roberta.encoder.layer.5.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="646" name="__module.roberta.encoder.layer.5.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9886,8 +9743,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="718" name="Constant_8916" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="234015516" size="12288" />
+		<layer id="647" name="Constant_8956" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="233607247" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -9896,7 +9753,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="719" name="__module.roberta.encoder.layer.5.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="648" name="__module.roberta.encoder.layer.5.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9911,14 +9768,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="497">
+				<port id="2" precision="FP32" names="446,input.97">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="720" name="__module.roberta.encoder.layer.5.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="649" name="__module.roberta.encoder.layer.5.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -9928,143 +9785,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="498">
+				<port id="1" precision="FP32" names="447,input.99">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="721" name="__module.roberta.encoder.layer.5.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="234027804" size="12288" />
+		<layer id="650" name="self.roberta.encoder.layer.5.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="233619535" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="722" name="__module.roberta.encoder.layer.5.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="651" name="Convert_64328" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="723" name="__module.roberta.encoder.layer.5.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="234040092" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="724" name="__module.roberta.encoder.layer.5.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="234040096" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="725" name="__module.roberta.encoder.layer.5.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="234040092" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="726" name="__module.roberta.encoder.layer.5.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="234040096" size="4" />
+		<layer id="652" name="self.roberta.encoder.layer.5.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="235978831" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="727" name="__module.roberta.encoder.layer.5.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="653" name="Convert_64331" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="728" name="Constant_1059155" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="234040100" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="729" name="Convert_1059156" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="654" name="self.roberta.encoder.layer.5.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="730" name="Constant_1059157" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="236399396" size="3072" />
+		<layer id="655" name="self.roberta.encoder.layer.5.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="235979599" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="731" name="__module.roberta.encoder.layer.5.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="656" name="self.roberta.encoder.layer.5.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="657" name="self.roberta.encoder.layer.5.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="732" name="__module.roberta.encoder.layer.5.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="658" name="__module.roberta.encoder.layer.5.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10085,8 +9923,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="733" name="Constant_8917" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="236402468" size="3072" />
+		<layer id="659" name="Constant_8957" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="235981135" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -10095,7 +9933,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="734" name="__module.roberta.encoder.layer.5.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="660" name="__module.roberta.encoder.layer.5.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10110,14 +9948,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="503,input.25">
+				<port id="2" precision="FP32" names="452,input.101">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="735" name="__module.roberta.encoder.layer.5.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="661" name="__module.roberta.encoder.layer.5.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10132,22 +9970,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="505">
+				<port id="2" precision="FP32" names="454,input.103">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="736" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="662" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="737" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="663" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10167,8 +10005,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="738" name="Constant_8918" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="236405540" size="3072" />
+		<layer id="664" name="Constant_8958" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="235984207" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -10177,7 +10015,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="739" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="665" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10199,8 +10037,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="740" name="Constant_8919" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="236408612" size="3072" />
+		<layer id="666" name="Constant_8959" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="235987279" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -10209,7 +10047,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="741" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="667" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10224,143 +10062,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="509,hidden_states.37">
+				<port id="2" precision="FP32" names="458,hidden_states.37">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="742" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="236411684" size="3072" />
+		<layer id="668" name="self.roberta.encoder.layer.6.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="235990351" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="743" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="669" name="Convert_219860" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="744" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="236414756" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="745" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="236414760" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="746" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="236414756" size="4" />
+		<layer id="670" name="self.roberta.encoder.layer.6.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="236580175" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="747" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="236414760" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="748" name="__module.roberta.encoder.layer.5.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="671" name="Convert_219863" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="749" name="Constant_1059159" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="236414764" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="672" name="self.roberta.encoder.layer.6.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="750" name="Convert_1059160" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="751" name="Constant_1059161" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="237004588" size="3072" />
+		<layer id="673" name="self.roberta.encoder.layer.6.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="236580943" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="752" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="674" name="self.roberta.encoder.layer.6.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="675" name="self.roberta.encoder.layer.6.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="753" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="676" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10381,8 +10200,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="754" name="Constant_8920" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="237007660" size="3072" />
+		<layer id="677" name="Constant_8960" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="236582479" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -10391,7 +10210,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="755" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="678" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10406,67 +10225,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="522,x.73">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="756" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="237010732" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="757" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="237010736" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="758" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="237010732" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="759" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="237010736" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="760" name="__module.roberta.encoder.layer.6.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="471">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="761" name="__module.roberta.encoder.layer.6.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="679" name="Constant_9281" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="762" name="__module.roberta.encoder.layer.6.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="680" name="__module.roberta.encoder.layer.6.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10479,7 +10253,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="526,x.75">
+				<port id="2" precision="FP32" names="473">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -10487,15 +10261,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="763" name="Constant_1677" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="681" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="527">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="764" name="__module.roberta.encoder.layer.6.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="682" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -10503,12 +10277,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="528">
+				<port id="2" precision="FP32" names="474,query_layer.13">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -10516,59 +10290,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="765" name="Constant_1059163" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="237010740" size="589824" />
+		<layer id="683" name="self.roberta.encoder.layer.6.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="236585551" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="766" name="Convert_1059164" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="684" name="Convert_215872" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="767" name="Constant_1059165" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="237600564" size="3072" />
+		<layer id="685" name="self.roberta.encoder.layer.6.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="237175375" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="768" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="686" name="Convert_215875" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="687" name="self.roberta.encoder.layer.6.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="688" name="self.roberta.encoder.layer.6.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="237176143" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="689" name="self.roberta.encoder.layer.6.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="690" name="self.roberta.encoder.layer.6.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="769" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="691" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10589,8 +10421,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="770" name="Constant_8921" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="237603636" size="3072" />
+		<layer id="692" name="Constant_8961" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="237177679" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -10599,7 +10431,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="771" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+		<layer id="693" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10614,67 +10446,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="531,x.77">
+				<port id="2" precision="FP32" names="477">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="772" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="237606708" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="773" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="237606712" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="774" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="237606708" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="775" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="237606712" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="776" name="__module.roberta.encoder.layer.6.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="777" name="__module.roberta.encoder.layer.6.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="694" name="Constant_9282" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="778" name="__module.roberta.encoder.layer.6.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="695" name="__module.roberta.encoder.layer.6.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10687,7 +10474,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="535,x.79">
+				<port id="2" precision="FP32" names="479">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -10695,15 +10482,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="779" name="Constant_1711" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="696" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="536">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="780" name="__module.roberta.encoder.layer.6.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="697" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -10711,12 +10498,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="537">
+				<port id="2" precision="FP32" names="480,key_layer.13">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -10724,59 +10511,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="781" name="Constant_1059167" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="237606716" size="589824" />
+		<layer id="698" name="self.roberta.encoder.layer.6.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="237180751" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="782" name="Convert_1059168" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="699" name="Convert_211884" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="783" name="Constant_1059169" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="238196540" size="3072" />
+		<layer id="700" name="self.roberta.encoder.layer.6.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="237770575" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="701" name="Convert_211887" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="784" name="__module.roberta.encoder.layer.6.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="702" name="self.roberta.encoder.layer.6.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="703" name="self.roberta.encoder.layer.6.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="237771343" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="704" name="self.roberta.encoder.layer.6.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="705" name="self.roberta.encoder.layer.6.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="785" name="__module.roberta.encoder.layer.6.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="706" name="__module.roberta.encoder.layer.6.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10797,8 +10642,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="786" name="Constant_8922" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="238199612" size="3072" />
+		<layer id="707" name="Constant_8962" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="237772879" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -10807,7 +10652,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="787" name="__module.roberta.encoder.layer.6.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="708" name="__module.roberta.encoder.layer.6.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10822,22 +10667,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="540,x.81">
+				<port id="2" precision="FP32" names="483">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="788" name="__module.roberta.encoder.layer.6.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="709" name="Constant_9283" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="789" name="__module.roberta.encoder.layer.6.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="710" name="__module.roberta.encoder.layer.6.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10850,7 +10695,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="544,x.83">
+				<port id="2" precision="FP32" names="485">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -10858,15 +10703,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="790" name="Constant_1745" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="711" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="545">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="791" name="__module.roberta.encoder.layer.6.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="712" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -10874,12 +10719,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="546">
+				<port id="2" precision="FP32" names="486,value_layer.13">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -10887,7 +10732,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="792" name="__module.roberta.encoder.layer.6.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="713" name="__module.roberta.encoder.layer.6.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10916,7 +10761,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="547,attn_output.25">
+				<port id="4" precision="FP32" names="487,attn_output.25">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -10924,15 +10769,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="793" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="714" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="794" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="715" name="__module.roberta.encoder.layer.6.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -10945,7 +10790,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="548,attn_output.27">
+				<port id="2" precision="FP32" names="488,attn_output.27">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -10953,15 +10798,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="795" name="Constant_9175" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="716" name="Constant_9284" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="796" name="__module.roberta.encoder.layer.6.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="717" name="__module.roberta.encoder.layer.6.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -10975,143 +10820,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="550">
+				<port id="2" precision="FP32" names="490,input.105">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="797" name="__module.roberta.encoder.layer.6.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="238202684" size="3072" />
+		<layer id="718" name="self.roberta.encoder.layer.6.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="237775951" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="798" name="__module.roberta.encoder.layer.6.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="719" name="Convert_223848" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="799" name="__module.roberta.encoder.layer.6.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="238205756" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="800" name="__module.roberta.encoder.layer.6.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="238205760" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="801" name="__module.roberta.encoder.layer.6.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="238205756" size="4" />
+		<layer id="720" name="self.roberta.encoder.layer.6.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="238365775" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="802" name="__module.roberta.encoder.layer.6.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="238205760" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="803" name="__module.roberta.encoder.layer.6.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="721" name="Convert_223851" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="804" name="Constant_1059171" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="238205764" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="722" name="self.roberta.encoder.layer.6.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="805" name="Convert_1059172" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="806" name="Constant_1059173" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="238795588" size="3072" />
+		<layer id="723" name="self.roberta.encoder.layer.6.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="238366543" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="807" name="__module.roberta.encoder.layer.6.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="724" name="self.roberta.encoder.layer.6.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="725" name="self.roberta.encoder.layer.6.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="808" name="__module.roberta.encoder.layer.6.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="726" name="__module.roberta.encoder.layer.6.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11132,8 +10958,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="809" name="Constant_8923" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="238798660" size="3072" />
+		<layer id="727" name="Constant_8963" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="238368079" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -11142,7 +10968,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="810" name="__module.roberta.encoder.layer.6.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="728" name="__module.roberta.encoder.layer.6.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11157,14 +10983,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="555,input.27">
+				<port id="2" precision="FP32" names="495,input.107">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="811" name="__module.roberta.encoder.layer.6.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="729" name="__module.roberta.encoder.layer.6.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11179,22 +11005,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="557">
+				<port id="2" precision="FP32" names="497,input.109">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="812" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="730" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="813" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="731" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11214,8 +11040,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="814" name="Constant_8924" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="238801732" size="3072" />
+		<layer id="732" name="Constant_8964" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="238371151" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -11224,7 +11050,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="815" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="733" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11246,8 +11072,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="816" name="Constant_8925" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="238804804" size="3072" />
+		<layer id="734" name="Constant_8965" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="238374223" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -11256,7 +11082,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="817" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="735" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11271,143 +11097,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="561,input_tensor.13">
+				<port id="2" precision="FP32" names="501,input.111">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="818" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="238807876" size="3072" />
+		<layer id="736" name="self.roberta.encoder.layer.6.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="238377295" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="819" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="737" name="Convert_68316" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="820" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="238810948" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="821" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="238810952" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="822" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="238810948" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="823" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="238810952" size="4" />
+		<layer id="738" name="self.roberta.encoder.layer.6.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="240736591" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="824" name="__module.roberta.encoder.layer.6.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="739" name="Convert_68319" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="825" name="Constant_1059175" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="238810956" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="826" name="Convert_1059176" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="740" name="self.roberta.encoder.layer.6.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="827" name="Constant_1059177" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="241170252" size="12288" />
+		<layer id="741" name="self.roberta.encoder.layer.6.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="240739663" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="828" name="__module.roberta.encoder.layer.6.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="742" name="self.roberta.encoder.layer.6.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="743" name="self.roberta.encoder.layer.6.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="829" name="__module.roberta.encoder.layer.6.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="744" name="__module.roberta.encoder.layer.6.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11428,8 +11235,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="830" name="Constant_8926" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="241182540" size="12288" />
+		<layer id="745" name="Constant_8966" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="240745807" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -11438,7 +11245,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="831" name="__module.roberta.encoder.layer.6.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="746" name="__module.roberta.encoder.layer.6.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11453,14 +11260,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="565">
+				<port id="2" precision="FP32" names="505,input.113">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="832" name="__module.roberta.encoder.layer.6.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="747" name="__module.roberta.encoder.layer.6.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11470,143 +11277,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="566">
+				<port id="1" precision="FP32" names="506,input.115">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="833" name="__module.roberta.encoder.layer.6.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="241194828" size="12288" />
+		<layer id="748" name="self.roberta.encoder.layer.6.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="240758095" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="834" name="__module.roberta.encoder.layer.6.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="749" name="Convert_72304" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="835" name="__module.roberta.encoder.layer.6.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="241207116" size="4" />
+		<layer id="750" name="self.roberta.encoder.layer.6.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="243117391" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="836" name="__module.roberta.encoder.layer.6.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="241207120" size="4" />
+		<layer id="751" name="Convert_72307" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="837" name="__module.roberta.encoder.layer.6.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="241207116" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="838" name="__module.roberta.encoder.layer.6.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="241207120" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="839" name="__module.roberta.encoder.layer.6.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="752" name="self.roberta.encoder.layer.6.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="840" name="Constant_1059179" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="241207124" size="2359296" />
+		<layer id="753" name="self.roberta.encoder.layer.6.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="243118159" size="1536" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="841" name="Convert_1059180" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="754" name="self.roberta.encoder.layer.6.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="842" name="Constant_1059181" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="243566420" size="3072" />
+			</input>
 			<output>
-				<port id="0" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
-					<dim>1</dim>
+					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="843" name="__module.roberta.encoder.layer.6.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="755" name="self.roberta.encoder.layer.6.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>768</dim>
-					<dim>1</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="844" name="__module.roberta.encoder.layer.6.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="756" name="__module.roberta.encoder.layer.6.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11627,8 +11415,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="845" name="Constant_8927" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="243569492" size="3072" />
+		<layer id="757" name="Constant_8967" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="243119695" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -11637,7 +11425,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="846" name="__module.roberta.encoder.layer.6.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="758" name="__module.roberta.encoder.layer.6.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11652,14 +11440,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="571,input.29">
+				<port id="2" precision="FP32" names="511,input.117">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="847" name="__module.roberta.encoder.layer.6.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="759" name="__module.roberta.encoder.layer.6.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11674,22 +11462,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="573">
+				<port id="2" precision="FP32" names="513,input.119">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="848" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="760" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="849" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="761" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11709,8 +11497,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="850" name="Constant_8928" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="243572564" size="3072" />
+		<layer id="762" name="Constant_8968" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="243122767" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -11719,7 +11507,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="851" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="763" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11741,8 +11529,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="852" name="Constant_8929" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="243575636" size="3072" />
+		<layer id="764" name="Constant_8969" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="243125839" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -11751,7 +11539,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="853" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="765" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11766,143 +11554,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="577,hidden_states.43">
+				<port id="2" precision="FP32" names="517,hidden_states.43">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="854" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="243578708" size="3072" />
+		<layer id="766" name="self.roberta.encoder.layer.7.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="243128911" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="855" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="767" name="Convert_235812" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="856" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="243581780" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="857" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="243581784" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="858" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="243581780" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="859" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="243581784" size="4" />
+		<layer id="768" name="self.roberta.encoder.layer.7.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="243718735" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="860" name="__module.roberta.encoder.layer.6.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="769" name="Convert_235815" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="861" name="Constant_1059183" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="243581788" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="770" name="self.roberta.encoder.layer.7.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="862" name="Convert_1059184" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="863" name="Constant_1059185" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="244171612" size="3072" />
+		<layer id="771" name="self.roberta.encoder.layer.7.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="243719503" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="864" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="772" name="self.roberta.encoder.layer.7.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="773" name="self.roberta.encoder.layer.7.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="865" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="774" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11923,8 +11692,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="866" name="Constant_8930" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="244174684" size="3072" />
+		<layer id="775" name="Constant_8970" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="243721039" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -11933,7 +11702,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="867" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="776" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -11948,67 +11717,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="590,x.85">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="868" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="244177756" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="869" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="244177760" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="870" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="244177756" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="871" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="244177760" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="872" name="__module.roberta.encoder.layer.7.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="530">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="873" name="__module.roberta.encoder.layer.7.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="777" name="Constant_9285" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="874" name="__module.roberta.encoder.layer.7.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="778" name="__module.roberta.encoder.layer.7.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12021,7 +11745,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="594,x.87">
+				<port id="2" precision="FP32" names="532">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -12029,15 +11753,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="875" name="Constant_1911" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="779" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="595">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="876" name="__module.roberta.encoder.layer.7.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="780" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -12045,12 +11769,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="596">
+				<port id="2" precision="FP32" names="533,query_layer.15">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -12058,91 +11782,118 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="877" name="Constant_1059187" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="244177764" size="589824" />
+		<layer id="781" name="self.roberta.encoder.layer.7.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="243724111" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="878" name="Convert_1059188" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="782" name="Convert_231824" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="879" name="Constant_1059189" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="244767588" size="3072" />
+		<layer id="783" name="self.roberta.encoder.layer.7.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="244313935" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="784" name="Convert_231827" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="880" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="785" name="self.roberta.encoder.layer.7.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="881" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+		<layer id="786" name="self.roberta.encoder.layer.7.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="244314703" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32">
+			</output>
+		</layer>
+		<layer id="787" name="self.roberta.encoder.layer.7.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="882" name="Constant_8931" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="244770660" size="3072" />
+		<layer id="788" name="self.roberta.encoder.layer.7.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="883" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/Add" type="Add" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="789" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -12150,73 +11901,59 @@
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="599,x.89">
+				<port id="2" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="884" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="244773732" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="885" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="244773736" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="886" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="244773732" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="887" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="244773736" size="4" />
+		<layer id="790" name="Constant_8971" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="244316239" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="888" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="791" name="__module.roberta.encoder.layer.7.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="536">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="889" name="__module.roberta.encoder.layer.7.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="792" name="Constant_9286" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="890" name="__module.roberta.encoder.layer.7.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="793" name="__module.roberta.encoder.layer.7.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12229,7 +11966,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="603,x.91">
+				<port id="2" precision="FP32" names="538">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -12237,15 +11974,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="891" name="Constant_1945" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="794" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="604">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="892" name="__module.roberta.encoder.layer.7.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="795" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -12253,12 +11990,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="605">
+				<port id="2" precision="FP32" names="539,key_layer.15">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -12266,90 +12003,148 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="893" name="Constant_1059191" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="244773740" size="589824" />
+		<layer id="796" name="self.roberta.encoder.layer.7.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="244319311" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="894" name="Convert_1059192" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="797" name="Convert_227836" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="895" name="Constant_1059193" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="245363564" size="3072" />
+		<layer id="798" name="self.roberta.encoder.layer.7.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="244909135" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="896" name="__module.roberta.encoder.layer.7.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="799" name="Convert_227839" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>768</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="897" name="__module.roberta.encoder.layer.7.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
+		<layer id="800" name="self.roberta.encoder.layer.7.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
 					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="898" name="Constant_8932" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="245366636" size="3072" />
+		<layer id="801" name="self.roberta.encoder.layer.7.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="244909903" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="802" name="self.roberta.encoder.layer.7.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="803" name="self.roberta.encoder.layer.7.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="804" name="__module.roberta.encoder.layer.7.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
+			<input>
+				<port id="0" precision="FP32">
+					<dim>-1</dim>
+					<dim>-1</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP32">
+					<dim>-1</dim>
+					<dim>-1</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="805" name="Constant_8972" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="244911439" size="3072" />
+			<output>
+				<port id="0" precision="FP32">
 					<dim>1</dim>
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="899" name="__module.roberta.encoder.layer.7.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="806" name="__module.roberta.encoder.layer.7.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12364,22 +12159,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="608,x.93">
+				<port id="2" precision="FP32" names="542">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="900" name="__module.roberta.encoder.layer.7.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="807" name="Constant_9287" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="901" name="__module.roberta.encoder.layer.7.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="808" name="__module.roberta.encoder.layer.7.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12392,7 +12187,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="612,x.95">
+				<port id="2" precision="FP32" names="544">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -12400,15 +12195,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="902" name="Constant_1979" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="809" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="613">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="903" name="__module.roberta.encoder.layer.7.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="810" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -12416,12 +12211,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="614">
+				<port id="2" precision="FP32" names="545,value_layer.15">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -12429,7 +12224,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="904" name="__module.roberta.encoder.layer.7.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="811" name="__module.roberta.encoder.layer.7.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12458,7 +12253,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="615,attn_output.29">
+				<port id="4" precision="FP32" names="546,attn_output.29">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -12466,15 +12261,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="905" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="812" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="906" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="813" name="__module.roberta.encoder.layer.7.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -12487,7 +12282,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="616,attn_output.31">
+				<port id="2" precision="FP32" names="547,attn_output.31">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -12495,15 +12290,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="907" name="Constant_9176" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="814" name="Constant_9288" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="908" name="__module.roberta.encoder.layer.7.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="815" name="__module.roberta.encoder.layer.7.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12517,143 +12312,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="618">
+				<port id="2" precision="FP32" names="549,input.121">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="909" name="__module.roberta.encoder.layer.7.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="245369708" size="3072" />
+		<layer id="816" name="self.roberta.encoder.layer.7.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="244914511" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="910" name="__module.roberta.encoder.layer.7.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="817" name="Convert_239800" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="911" name="__module.roberta.encoder.layer.7.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="245372780" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="912" name="__module.roberta.encoder.layer.7.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="245372784" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="913" name="__module.roberta.encoder.layer.7.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="245372780" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="914" name="__module.roberta.encoder.layer.7.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="245372784" size="4" />
+		<layer id="818" name="self.roberta.encoder.layer.7.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="245504335" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="915" name="__module.roberta.encoder.layer.7.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="819" name="Convert_239803" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="916" name="Constant_1059195" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="245372788" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="820" name="self.roberta.encoder.layer.7.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="917" name="Convert_1059196" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="918" name="Constant_1059197" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="245962612" size="3072" />
+		<layer id="821" name="self.roberta.encoder.layer.7.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="245505103" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="919" name="__module.roberta.encoder.layer.7.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="822" name="self.roberta.encoder.layer.7.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="823" name="self.roberta.encoder.layer.7.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="920" name="__module.roberta.encoder.layer.7.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="824" name="__module.roberta.encoder.layer.7.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12674,8 +12450,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="921" name="Constant_8933" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="245965684" size="3072" />
+		<layer id="825" name="Constant_8973" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="245506639" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -12684,7 +12460,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="922" name="__module.roberta.encoder.layer.7.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="826" name="__module.roberta.encoder.layer.7.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12699,14 +12475,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="623,input.31">
+				<port id="2" precision="FP32" names="554,input.123">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="923" name="__module.roberta.encoder.layer.7.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="827" name="__module.roberta.encoder.layer.7.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12721,22 +12497,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="625">
+				<port id="2" precision="FP32" names="556,input.125">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="924" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="828" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="925" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="829" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12756,8 +12532,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="926" name="Constant_8934" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="245968756" size="3072" />
+		<layer id="830" name="Constant_8974" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="245509711" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -12766,7 +12542,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="927" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="831" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12788,8 +12564,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="928" name="Constant_8935" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="245971828" size="3072" />
+		<layer id="832" name="Constant_8975" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="245512783" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -12798,7 +12574,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="929" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="833" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12813,143 +12589,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="629,input_tensor.15">
+				<port id="2" precision="FP32" names="560,input.127">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="930" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="245974900" size="3072" />
+		<layer id="834" name="self.roberta.encoder.layer.7.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="245515855" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="931" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="835" name="Convert_76292" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="932" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="245977972" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="933" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="245977976" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="934" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="245977972" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="935" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="245977976" size="4" />
+		<layer id="836" name="self.roberta.encoder.layer.7.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="247875151" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="936" name="__module.roberta.encoder.layer.7.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="837" name="Convert_76295" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="937" name="Constant_1059199" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="245977980" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="938" name="Convert_1059200" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="838" name="self.roberta.encoder.layer.7.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="939" name="Constant_1059201" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="248337276" size="12288" />
+		<layer id="839" name="self.roberta.encoder.layer.7.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="247878223" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="940" name="__module.roberta.encoder.layer.7.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="840" name="self.roberta.encoder.layer.7.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="841" name="self.roberta.encoder.layer.7.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="941" name="__module.roberta.encoder.layer.7.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="842" name="__module.roberta.encoder.layer.7.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12970,8 +12727,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="942" name="Constant_8936" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="248349564" size="12288" />
+		<layer id="843" name="Constant_8976" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="247884367" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -12980,7 +12737,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="943" name="__module.roberta.encoder.layer.7.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="844" name="__module.roberta.encoder.layer.7.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -12995,14 +12752,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="633">
+				<port id="2" precision="FP32" names="564,input.129">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="944" name="__module.roberta.encoder.layer.7.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="845" name="__module.roberta.encoder.layer.7.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13012,143 +12769,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="634">
+				<port id="1" precision="FP32" names="565,input.131">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="945" name="__module.roberta.encoder.layer.7.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="248361852" size="12288" />
+		<layer id="846" name="self.roberta.encoder.layer.7.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="247896655" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="946" name="__module.roberta.encoder.layer.7.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="847" name="Convert_80280" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="947" name="__module.roberta.encoder.layer.7.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="248374140" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="948" name="__module.roberta.encoder.layer.7.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="248374144" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="949" name="__module.roberta.encoder.layer.7.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="248374140" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="950" name="__module.roberta.encoder.layer.7.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="248374144" size="4" />
+		<layer id="848" name="self.roberta.encoder.layer.7.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="250255951" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="951" name="__module.roberta.encoder.layer.7.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="849" name="Convert_80283" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="952" name="Constant_1059203" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="248374148" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="953" name="Convert_1059204" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="850" name="self.roberta.encoder.layer.7.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="954" name="Constant_1059205" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="250733444" size="3072" />
+		<layer id="851" name="self.roberta.encoder.layer.7.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="250256719" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="955" name="__module.roberta.encoder.layer.7.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="852" name="self.roberta.encoder.layer.7.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="853" name="self.roberta.encoder.layer.7.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="956" name="__module.roberta.encoder.layer.7.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="854" name="__module.roberta.encoder.layer.7.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13169,8 +12907,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="957" name="Constant_8937" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="250736516" size="3072" />
+		<layer id="855" name="Constant_8977" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="250258255" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -13179,7 +12917,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="958" name="__module.roberta.encoder.layer.7.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="856" name="__module.roberta.encoder.layer.7.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13194,14 +12932,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="639,input.33">
+				<port id="2" precision="FP32" names="570,input.133">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="959" name="__module.roberta.encoder.layer.7.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="857" name="__module.roberta.encoder.layer.7.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13216,22 +12954,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="641">
+				<port id="2" precision="FP32" names="572,input.135">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="960" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="858" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="961" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="859" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13251,8 +12989,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="962" name="Constant_8938" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="250739588" size="3072" />
+		<layer id="860" name="Constant_8978" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="250261327" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -13261,7 +12999,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="963" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="861" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13283,8 +13021,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="964" name="Constant_8939" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="250742660" size="3072" />
+		<layer id="862" name="Constant_8979" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="250264399" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -13293,7 +13031,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="965" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="863" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13308,143 +13046,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="645,hidden_states.49">
+				<port id="2" precision="FP32" names="576,hidden_states.49">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="966" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="250745732" size="3072" />
+		<layer id="864" name="self.roberta.encoder.layer.8.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="250267471" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="967" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="865" name="Convert_251764" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="968" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="250748804" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="969" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="250748808" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="970" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="250748804" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="971" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="250748808" size="4" />
+		<layer id="866" name="self.roberta.encoder.layer.8.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="250857295" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="972" name="__module.roberta.encoder.layer.7.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="867" name="Convert_251767" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="973" name="Constant_1059207" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="250748812" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="868" name="self.roberta.encoder.layer.8.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="974" name="Convert_1059208" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="975" name="Constant_1059209" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="251338636" size="3072" />
+		<layer id="869" name="self.roberta.encoder.layer.8.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="250858063" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="976" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="870" name="self.roberta.encoder.layer.8.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="871" name="self.roberta.encoder.layer.8.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="977" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="872" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13465,8 +13184,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="978" name="Constant_8940" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="251341708" size="3072" />
+		<layer id="873" name="Constant_8980" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="250859599" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -13475,7 +13194,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="979" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="874" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13490,67 +13209,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="658,x.97">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="980" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="251344780" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="981" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="251344784" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="982" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="251344780" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="983" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="251344784" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="984" name="__module.roberta.encoder.layer.8.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="589">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="985" name="__module.roberta.encoder.layer.8.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="875" name="Constant_9289" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="986" name="__module.roberta.encoder.layer.8.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="876" name="__module.roberta.encoder.layer.8.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13563,7 +13237,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="662,x.99">
+				<port id="2" precision="FP32" names="591">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -13571,15 +13245,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="987" name="Constant_2145" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="877" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="663">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="988" name="__module.roberta.encoder.layer.8.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="878" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -13587,12 +13261,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="664">
+				<port id="2" precision="FP32" names="592,query_layer.17">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -13600,91 +13274,118 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="989" name="Constant_1059211" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="251344788" size="589824" />
+		<layer id="879" name="self.roberta.encoder.layer.8.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="250862671" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="990" name="Convert_1059212" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="880" name="Convert_247776" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="991" name="Constant_1059213" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="251934612" size="3072" />
+		<layer id="881" name="self.roberta.encoder.layer.8.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="251452495" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="882" name="Convert_247779" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="992" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="883" name="self.roberta.encoder.layer.8.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="993" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+		<layer id="884" name="self.roberta.encoder.layer.8.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="251453263" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32">
+			</output>
+		</layer>
+		<layer id="885" name="self.roberta.encoder.layer.8.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="994" name="Constant_8941" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="251937684" size="3072" />
+		<layer id="886" name="self.roberta.encoder.layer.8.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="995" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/Add" type="Add" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="887" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -13692,73 +13393,59 @@
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="667,x.101">
+				<port id="2" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="996" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="251940756" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="997" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="251940760" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="998" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="251940756" size="4" />
+		<layer id="888" name="Constant_8981" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="251454799" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="999" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="251940760" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1000" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="889" name="__module.roberta.encoder.layer.8.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="595">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1001" name="__module.roberta.encoder.layer.8.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="890" name="Constant_9290" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1002" name="__module.roberta.encoder.layer.8.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="891" name="__module.roberta.encoder.layer.8.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13771,7 +13458,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="671,x.103">
+				<port id="2" precision="FP32" names="597">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -13779,15 +13466,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1003" name="Constant_2179" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="892" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="672">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1004" name="__module.roberta.encoder.layer.8.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="893" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -13795,12 +13482,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="673">
+				<port id="2" precision="FP32" names="598,key_layer.17">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -13808,59 +13495,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1005" name="Constant_1059215" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="251940764" size="589824" />
+		<layer id="894" name="self.roberta.encoder.layer.8.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="251457871" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1006" name="Convert_1059216" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="895" name="Convert_243788" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1007" name="Constant_1059217" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="252530588" size="3072" />
+		<layer id="896" name="self.roberta.encoder.layer.8.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="252047695" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="897" name="Convert_243791" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1008" name="__module.roberta.encoder.layer.8.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="898" name="self.roberta.encoder.layer.8.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="899" name="self.roberta.encoder.layer.8.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="252048463" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="900" name="self.roberta.encoder.layer.8.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="901" name="self.roberta.encoder.layer.8.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1009" name="__module.roberta.encoder.layer.8.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="902" name="__module.roberta.encoder.layer.8.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13881,8 +13626,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1010" name="Constant_8942" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="252533660" size="3072" />
+		<layer id="903" name="Constant_8982" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="252049999" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -13891,7 +13636,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1011" name="__module.roberta.encoder.layer.8.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="904" name="__module.roberta.encoder.layer.8.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13906,22 +13651,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="676,x.105">
+				<port id="2" precision="FP32" names="601">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1012" name="__module.roberta.encoder.layer.8.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="905" name="Constant_9291" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1013" name="__module.roberta.encoder.layer.8.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="906" name="__module.roberta.encoder.layer.8.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -13934,7 +13679,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="680,x.107">
+				<port id="2" precision="FP32" names="603">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -13942,15 +13687,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1014" name="Constant_2213" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="907" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="681">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1015" name="__module.roberta.encoder.layer.8.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="908" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -13958,12 +13703,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="682">
+				<port id="2" precision="FP32" names="604,value_layer.17">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -13971,7 +13716,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1016" name="__module.roberta.encoder.layer.8.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="909" name="__module.roberta.encoder.layer.8.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14000,7 +13745,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="683,attn_output.33">
+				<port id="4" precision="FP32" names="605,attn_output.33">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -14008,15 +13753,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1017" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="910" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1018" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="911" name="__module.roberta.encoder.layer.8.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -14029,7 +13774,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="684,attn_output.35">
+				<port id="2" precision="FP32" names="606,attn_output.35">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -14037,15 +13782,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1019" name="Constant_9177" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="912" name="Constant_9292" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1020" name="__module.roberta.encoder.layer.8.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="913" name="__module.roberta.encoder.layer.8.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14059,143 +13804,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="686">
+				<port id="2" precision="FP32" names="608,input.137">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1021" name="__module.roberta.encoder.layer.8.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="252536732" size="3072" />
+		<layer id="914" name="self.roberta.encoder.layer.8.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="252053071" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1022" name="__module.roberta.encoder.layer.8.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="915" name="Convert_255752" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1023" name="__module.roberta.encoder.layer.8.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="252539804" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1024" name="__module.roberta.encoder.layer.8.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="252539808" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1025" name="__module.roberta.encoder.layer.8.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="252539804" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1026" name="__module.roberta.encoder.layer.8.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="252539808" size="4" />
+		<layer id="916" name="self.roberta.encoder.layer.8.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="252642895" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1027" name="__module.roberta.encoder.layer.8.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="917" name="Convert_255755" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1028" name="Constant_1059219" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="252539812" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="918" name="self.roberta.encoder.layer.8.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="1029" name="Convert_1059220" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1030" name="Constant_1059221" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="253129636" size="3072" />
+		<layer id="919" name="self.roberta.encoder.layer.8.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="252643663" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1031" name="__module.roberta.encoder.layer.8.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="920" name="self.roberta.encoder.layer.8.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1032" name="__module.roberta.encoder.layer.8.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="921" name="self.roberta.encoder.layer.8.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="922" name="__module.roberta.encoder.layer.8.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14216,8 +13942,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1033" name="Constant_8943" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="253132708" size="3072" />
+		<layer id="923" name="Constant_8983" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="252645199" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -14226,7 +13952,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1034" name="__module.roberta.encoder.layer.8.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="924" name="__module.roberta.encoder.layer.8.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14241,14 +13967,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="691,input.35">
+				<port id="2" precision="FP32" names="613,input.139">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1035" name="__module.roberta.encoder.layer.8.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="925" name="__module.roberta.encoder.layer.8.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14263,22 +13989,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="693">
+				<port id="2" precision="FP32" names="615,input.141">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1036" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="926" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1037" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="927" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14298,8 +14024,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1038" name="Constant_8944" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="253135780" size="3072" />
+		<layer id="928" name="Constant_8984" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="252648271" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -14308,7 +14034,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1039" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="929" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14330,8 +14056,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1040" name="Constant_8945" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="253138852" size="3072" />
+		<layer id="930" name="Constant_8985" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="252651343" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -14340,7 +14066,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1041" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="931" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14355,143 +14081,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="697,input_tensor.17">
+				<port id="2" precision="FP32" names="619,input.143">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1042" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="253141924" size="3072" />
+		<layer id="932" name="self.roberta.encoder.layer.8.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="252654415" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1043" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="933" name="Convert_84268" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1044" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="253144996" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1045" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="253145000" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1046" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="253144996" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1047" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="253145000" size="4" />
+		<layer id="934" name="self.roberta.encoder.layer.8.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="255013711" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1048" name="__module.roberta.encoder.layer.8.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="935" name="Convert_84271" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1049" name="Constant_1059223" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="253145004" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1050" name="Convert_1059224" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="936" name="self.roberta.encoder.layer.8.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1051" name="Constant_1059225" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="255504300" size="12288" />
+		<layer id="937" name="self.roberta.encoder.layer.8.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="255016783" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1052" name="__module.roberta.encoder.layer.8.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="938" name="self.roberta.encoder.layer.8.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="939" name="self.roberta.encoder.layer.8.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1053" name="__module.roberta.encoder.layer.8.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="940" name="__module.roberta.encoder.layer.8.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14512,8 +14219,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1054" name="Constant_8946" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="255516588" size="12288" />
+		<layer id="941" name="Constant_8986" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="255022927" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -14522,7 +14229,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1055" name="__module.roberta.encoder.layer.8.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="942" name="__module.roberta.encoder.layer.8.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14537,14 +14244,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="701">
+				<port id="2" precision="FP32" names="623,input.145">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1056" name="__module.roberta.encoder.layer.8.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="943" name="__module.roberta.encoder.layer.8.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14554,143 +14261,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="702">
+				<port id="1" precision="FP32" names="624,input.147">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1057" name="__module.roberta.encoder.layer.8.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="255528876" size="12288" />
+		<layer id="944" name="self.roberta.encoder.layer.8.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="255035215" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1058" name="__module.roberta.encoder.layer.8.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="945" name="Convert_88256" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1059" name="__module.roberta.encoder.layer.8.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="255541164" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1060" name="__module.roberta.encoder.layer.8.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="255541168" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1061" name="__module.roberta.encoder.layer.8.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="255541164" size="4" />
+		<layer id="946" name="self.roberta.encoder.layer.8.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="257394511" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1062" name="__module.roberta.encoder.layer.8.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="255541168" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1063" name="__module.roberta.encoder.layer.8.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="947" name="Convert_88259" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1064" name="Constant_1059227" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="255541172" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1065" name="Convert_1059228" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="948" name="self.roberta.encoder.layer.8.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1066" name="Constant_1059229" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="257900468" size="3072" />
+		<layer id="949" name="self.roberta.encoder.layer.8.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="257395279" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1067" name="__module.roberta.encoder.layer.8.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="950" name="self.roberta.encoder.layer.8.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="951" name="self.roberta.encoder.layer.8.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1068" name="__module.roberta.encoder.layer.8.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="952" name="__module.roberta.encoder.layer.8.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14711,8 +14399,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1069" name="Constant_8947" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="257903540" size="3072" />
+		<layer id="953" name="Constant_8987" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="257396815" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -14721,7 +14409,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1070" name="__module.roberta.encoder.layer.8.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="954" name="__module.roberta.encoder.layer.8.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14736,14 +14424,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="707,input.37">
+				<port id="2" precision="FP32" names="629,input.149">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1071" name="__module.roberta.encoder.layer.8.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="955" name="__module.roberta.encoder.layer.8.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14758,22 +14446,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="709">
+				<port id="2" precision="FP32" names="631,input.151">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1072" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="956" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1073" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="957" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14793,8 +14481,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1074" name="Constant_8948" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="257906612" size="3072" />
+		<layer id="958" name="Constant_8988" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="257399887" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -14803,7 +14491,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1075" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="959" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14825,8 +14513,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1076" name="Constant_8949" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="257909684" size="3072" />
+		<layer id="960" name="Constant_8989" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="257402959" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -14835,7 +14523,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1077" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="961" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -14850,143 +14538,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="713,hidden_states.55">
+				<port id="2" precision="FP32" names="635,hidden_states.55">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1078" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="257912756" size="3072" />
+		<layer id="962" name="self.roberta.encoder.layer.9.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="257406031" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1079" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="963" name="Convert_267716" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1080" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="257915828" size="4" />
+		<layer id="964" name="self.roberta.encoder.layer.9.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="257995855" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1081" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="257915832" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1082" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="257915828" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1083" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="257915832" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1084" name="__module.roberta.encoder.layer.8.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="965" name="Convert_267719" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1085" name="Constant_1059231" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="257915836" size="589824" />
-			<output>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1086" name="Convert_1059232" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="966" name="self.roberta.encoder.layer.9.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1087" name="Constant_1059233" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="258505660" size="3072" />
+		<layer id="967" name="self.roberta.encoder.layer.9.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="257996623" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1088" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="968" name="self.roberta.encoder.layer.9.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="969" name="self.roberta.encoder.layer.9.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1089" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="970" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15007,8 +14676,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1090" name="Constant_8950" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="258508732" size="3072" />
+		<layer id="971" name="Constant_8990" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="257998159" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -15017,7 +14686,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1091" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="972" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15032,67 +14701,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="726,x.109">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1092" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="258511804" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1093" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="258511808" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1094" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="258511804" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1095" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="258511808" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1096" name="__module.roberta.encoder.layer.9.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="648">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1097" name="__module.roberta.encoder.layer.9.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="973" name="Constant_9293" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1098" name="__module.roberta.encoder.layer.9.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="974" name="__module.roberta.encoder.layer.9.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15105,7 +14729,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="730,x.111">
+				<port id="2" precision="FP32" names="650">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -15113,15 +14737,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1099" name="Constant_2379" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="975" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="731">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1100" name="__module.roberta.encoder.layer.9.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="976" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -15129,12 +14753,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="732">
+				<port id="2" precision="FP32" names="651,query_layer.19">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -15142,91 +14766,118 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1101" name="Constant_1059235" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="258511812" size="589824" />
+		<layer id="977" name="self.roberta.encoder.layer.9.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="258001231" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1102" name="Convert_1059236" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="978" name="Convert_263728" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1103" name="Constant_1059237" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="259101636" size="3072" />
+		<layer id="979" name="self.roberta.encoder.layer.9.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="258591055" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="980" name="Convert_263731" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1104" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="981" name="self.roberta.encoder.layer.9.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1105" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+		<layer id="982" name="self.roberta.encoder.layer.9.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="258591823" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32">
+			</output>
+		</layer>
+		<layer id="983" name="self.roberta.encoder.layer.9.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1106" name="Constant_8951" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="259104708" size="3072" />
+		<layer id="984" name="self.roberta.encoder.layer.9.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1107" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/Add" type="Add" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="985" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -15234,73 +14885,59 @@
 					<dim>768</dim>
 				</port>
 				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="735,x.113">
+				<port id="2" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1108" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="259107780" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1109" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="259107784" size="4" />
+		<layer id="986" name="Constant_8991" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="258593359" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1110" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="259107780" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1111" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="259107784" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1112" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="987" name="__module.roberta.encoder.layer.9.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="654">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1113" name="__module.roberta.encoder.layer.9.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="988" name="Constant_9294" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1114" name="__module.roberta.encoder.layer.9.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="989" name="__module.roberta.encoder.layer.9.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15313,7 +14950,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="739,x.115">
+				<port id="2" precision="FP32" names="656">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -15321,15 +14958,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1115" name="Constant_2413" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="990" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="740">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1116" name="__module.roberta.encoder.layer.9.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="991" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -15337,12 +14974,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="741">
+				<port id="2" precision="FP32" names="657,key_layer.19">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -15350,59 +14987,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1117" name="Constant_1059239" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="259107788" size="589824" />
+		<layer id="992" name="self.roberta.encoder.layer.9.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="258596431" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1118" name="Convert_1059240" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="993" name="Convert_259740" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1119" name="Constant_1059241" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="259697612" size="3072" />
+		<layer id="994" name="self.roberta.encoder.layer.9.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="259186255" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="995" name="Convert_259743" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1120" name="__module.roberta.encoder.layer.9.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="996" name="self.roberta.encoder.layer.9.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="997" name="self.roberta.encoder.layer.9.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="259187023" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="998" name="self.roberta.encoder.layer.9.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="999" name="self.roberta.encoder.layer.9.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1121" name="__module.roberta.encoder.layer.9.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1000" name="__module.roberta.encoder.layer.9.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15423,8 +15118,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1122" name="Constant_8952" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="259700684" size="3072" />
+		<layer id="1001" name="Constant_8992" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="259188559" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -15433,7 +15128,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1123" name="__module.roberta.encoder.layer.9.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1002" name="__module.roberta.encoder.layer.9.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15448,22 +15143,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="744,x.117">
+				<port id="2" precision="FP32" names="660">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1124" name="__module.roberta.encoder.layer.9.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="1003" name="Constant_9295" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1125" name="__module.roberta.encoder.layer.9.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="1004" name="__module.roberta.encoder.layer.9.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15476,7 +15171,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="748,x.119">
+				<port id="2" precision="FP32" names="662">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -15484,15 +15179,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1126" name="Constant_2447" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="1005" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="749">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1127" name="__module.roberta.encoder.layer.9.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="1006" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -15500,12 +15195,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="750">
+				<port id="2" precision="FP32" names="663,value_layer.19">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -15513,7 +15208,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1128" name="__module.roberta.encoder.layer.9.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="1007" name="__module.roberta.encoder.layer.9.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15542,7 +15237,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="751,attn_output.37">
+				<port id="4" precision="FP32" names="664,attn_output.37">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -15550,15 +15245,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1129" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="1008" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1130" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="1009" name="__module.roberta.encoder.layer.9.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -15571,7 +15266,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="752,attn_output.39">
+				<port id="2" precision="FP32" names="665,attn_output.39">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -15579,15 +15274,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1131" name="Constant_9178" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="1010" name="Constant_9296" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1132" name="__module.roberta.encoder.layer.9.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="1011" name="__module.roberta.encoder.layer.9.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15601,165 +15296,146 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="754">
+				<port id="2" precision="FP32" names="667,input.153">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1133" name="__module.roberta.encoder.layer.9.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="259703756" size="3072" />
+		<layer id="1012" name="self.roberta.encoder.layer.9.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="259191631" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1134" name="__module.roberta.encoder.layer.9.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1013" name="Convert_271704" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1135" name="__module.roberta.encoder.layer.9.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="259706828" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1136" name="__module.roberta.encoder.layer.9.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="259706832" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1137" name="__module.roberta.encoder.layer.9.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="259706828" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1138" name="__module.roberta.encoder.layer.9.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="259706832" size="4" />
+		<layer id="1014" name="self.roberta.encoder.layer.9.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="259781455" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1139" name="__module.roberta.encoder.layer.9.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1015" name="Convert_271707" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1140" name="Constant_1059243" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="259706836" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="1016" name="self.roberta.encoder.layer.9.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="1141" name="Convert_1059244" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1142" name="Constant_1059245" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="260296660" size="3072" />
+		<layer id="1017" name="self.roberta.encoder.layer.9.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="259782223" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1143" name="__module.roberta.encoder.layer.9.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1018" name="self.roberta.encoder.layer.9.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1144" name="__module.roberta.encoder.layer.9.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
+		<layer id="1019" name="self.roberta.encoder.layer.9.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1020" name="__module.roberta.encoder.layer.9.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
+			<input>
+				<port id="0" precision="FP32">
+					<dim>-1</dim>
+					<dim>-1</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP32">
+					<dim>-1</dim>
+					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1145" name="Constant_8953" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="260299732" size="3072" />
+		<layer id="1021" name="Constant_8993" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="259783759" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -15768,7 +15444,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1146" name="__module.roberta.encoder.layer.9.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1022" name="__module.roberta.encoder.layer.9.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15783,14 +15459,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="759,input.39">
+				<port id="2" precision="FP32" names="672,input.155">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1147" name="__module.roberta.encoder.layer.9.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="1023" name="__module.roberta.encoder.layer.9.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15805,22 +15481,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="761">
+				<port id="2" precision="FP32" names="674,input.157">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1148" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="1024" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1149" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="1025" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15840,8 +15516,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1150" name="Constant_8954" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="260302804" size="3072" />
+		<layer id="1026" name="Constant_8994" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="259786831" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -15850,7 +15526,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1151" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="1027" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15872,8 +15548,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1152" name="Constant_8955" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="260305876" size="3072" />
+		<layer id="1028" name="Constant_8995" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="259789903" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -15882,7 +15558,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1153" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="1029" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -15897,143 +15573,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="765,input_tensor.19">
+				<port id="2" precision="FP32" names="678,input.159">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1154" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="260308948" size="3072" />
+		<layer id="1030" name="self.roberta.encoder.layer.9.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="259792975" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1155" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1031" name="Convert_92244" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1156" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="260312020" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1157" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="260312024" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1158" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="260312020" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1159" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="260312024" size="4" />
+		<layer id="1032" name="self.roberta.encoder.layer.9.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="262152271" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1160" name="__module.roberta.encoder.layer.9.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1033" name="Convert_92247" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1161" name="Constant_1059247" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="260312028" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1162" name="Convert_1059248" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1034" name="self.roberta.encoder.layer.9.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1163" name="Constant_1059249" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="262671324" size="12288" />
+		<layer id="1035" name="self.roberta.encoder.layer.9.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="262155343" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1164" name="__module.roberta.encoder.layer.9.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1036" name="self.roberta.encoder.layer.9.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1037" name="self.roberta.encoder.layer.9.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1165" name="__module.roberta.encoder.layer.9.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1038" name="__module.roberta.encoder.layer.9.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16054,8 +15711,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1166" name="Constant_8956" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="262683612" size="12288" />
+		<layer id="1039" name="Constant_8996" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="262161487" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -16064,7 +15721,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1167" name="__module.roberta.encoder.layer.9.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1040" name="__module.roberta.encoder.layer.9.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16079,14 +15736,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="769">
+				<port id="2" precision="FP32" names="682,input.161">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1168" name="__module.roberta.encoder.layer.9.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="1041" name="__module.roberta.encoder.layer.9.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16096,143 +15753,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="770">
+				<port id="1" precision="FP32" names="683,input.163">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1169" name="__module.roberta.encoder.layer.9.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="262695900" size="12288" />
+		<layer id="1042" name="self.roberta.encoder.layer.9.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="262173775" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1170" name="__module.roberta.encoder.layer.9.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1043" name="Convert_96232" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1171" name="__module.roberta.encoder.layer.9.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="262708188" size="4" />
+		<layer id="1044" name="self.roberta.encoder.layer.9.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="264533071" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1172" name="__module.roberta.encoder.layer.9.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="262708192" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1173" name="__module.roberta.encoder.layer.9.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="262708188" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1174" name="__module.roberta.encoder.layer.9.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="262708192" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1175" name="__module.roberta.encoder.layer.9.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1045" name="Convert_96235" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1176" name="Constant_1059251" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="262708196" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1177" name="Convert_1059252" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1046" name="self.roberta.encoder.layer.9.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1178" name="Constant_1059253" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="265067492" size="3072" />
+		<layer id="1047" name="self.roberta.encoder.layer.9.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="264533839" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1179" name="__module.roberta.encoder.layer.9.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1048" name="self.roberta.encoder.layer.9.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1049" name="self.roberta.encoder.layer.9.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1180" name="__module.roberta.encoder.layer.9.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1050" name="__module.roberta.encoder.layer.9.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16253,8 +15891,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1181" name="Constant_8957" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="265070564" size="3072" />
+		<layer id="1051" name="Constant_8997" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="264535375" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -16263,7 +15901,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1182" name="__module.roberta.encoder.layer.9.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1052" name="__module.roberta.encoder.layer.9.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16278,14 +15916,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="775,input.41">
+				<port id="2" precision="FP32" names="688,input.165">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1183" name="__module.roberta.encoder.layer.9.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="1053" name="__module.roberta.encoder.layer.9.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16300,22 +15938,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="777">
+				<port id="2" precision="FP32" names="690,input.167">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1184" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="1054" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1185" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="1055" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16335,8 +15973,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1186" name="Constant_8958" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="265073636" size="3072" />
+		<layer id="1056" name="Constant_8998" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="264538447" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -16345,7 +15983,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1187" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="1057" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16367,8 +16005,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1188" name="Constant_8959" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="265076708" size="3072" />
+		<layer id="1058" name="Constant_8999" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="264541519" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -16377,7 +16015,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1189" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="1059" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16392,143 +16030,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="781,hidden_states.61">
+				<port id="2" precision="FP32" names="694,hidden_states.61">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1190" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="265079780" size="3072" />
+		<layer id="1060" name="self.roberta.encoder.layer.10.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="264544591" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1191" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1061" name="Convert_283668" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1192" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="265082852" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1193" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="265082856" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1194" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="265082852" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1195" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="265082856" size="4" />
+		<layer id="1062" name="self.roberta.encoder.layer.10.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="265134415" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1196" name="__module.roberta.encoder.layer.9.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1063" name="Convert_283671" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1197" name="Constant_1059255" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="265082860" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="1064" name="self.roberta.encoder.layer.10.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="1198" name="Convert_1059256" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1199" name="Constant_1059257" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="265672684" size="3072" />
+		<layer id="1065" name="self.roberta.encoder.layer.10.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="265135183" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1200" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1066" name="self.roberta.encoder.layer.10.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1067" name="self.roberta.encoder.layer.10.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1201" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1068" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16549,8 +16168,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1202" name="Constant_8960" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="265675756" size="3072" />
+		<layer id="1069" name="Constant_9000" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="265136719" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -16559,7 +16178,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1203" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1070" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16574,169 +16193,182 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="794,x.121">
+				<port id="2" precision="FP32" names="707">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1204" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="265678828" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1205" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="265678832" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1206" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="265678828" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1207" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="265678832" size="4" />
+		<layer id="1071" name="Constant_9297" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="I64">
+					<dim>4</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1208" name="__module.roberta.encoder.layer.10.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1072" name="__module.roberta.encoder.layer.10.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="I64">
+					<dim>4</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="709">
 					<dim>-1</dim>
 					<dim>-1</dim>
-					<dim>768</dim>
+					<dim>12</dim>
+					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1209" name="__module.roberta.encoder.layer.10.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="1073" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1210" name="__module.roberta.encoder.layer.10.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
-			<data special_zero="true" />
+		<layer id="1074" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
-					<dim>768</dim>
+					<dim>12</dim>
+					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="798,x.123">
-					<dim>-1</dim>
+				<port id="2" precision="FP32" names="710,query_layer.21">
 					<dim>-1</dim>
 					<dim>12</dim>
+					<dim>-1</dim>
 					<dim>64</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1211" name="Constant_2613" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="1075" name="self.roberta.encoder.layer.10.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="265139791" size="589824" />
 			<output>
-				<port id="0" precision="I64" names="799">
-					<dim>4</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1212" name="__module.roberta.encoder.layer.10.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="1076" name="Convert_279680" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>12</dim>
-					<dim>64</dim>
-				</port>
-				<port id="1" precision="I64">
-					<dim>4</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="800">
-					<dim>-1</dim>
-					<dim>12</dim>
-					<dim>-1</dim>
-					<dim>64</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1213" name="Constant_1059259" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="265678836" size="589824" />
+		<layer id="1077" name="self.roberta.encoder.layer.10.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="265729615" size="768" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1078" name="Convert_279683" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1214" name="Convert_1059260" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1079" name="self.roberta.encoder.layer.10.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1215" name="Constant_1059261" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="266268660" size="3072" />
+		<layer id="1080" name="self.roberta.encoder.layer.10.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="265730383" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1216" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1081" name="self.roberta.encoder.layer.10.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1082" name="self.roberta.encoder.layer.10.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1217" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1083" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16757,8 +16389,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1218" name="Constant_8961" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="266271732" size="3072" />
+		<layer id="1084" name="Constant_9001" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="265731919" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -16767,7 +16399,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1219" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1085" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16782,67 +16414,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="803,x.125">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1220" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="266274804" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1221" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="266274808" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1222" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="266274804" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1223" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="266274808" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1224" name="__module.roberta.encoder.layer.10.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="713">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1225" name="__module.roberta.encoder.layer.10.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="1086" name="Constant_9298" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1226" name="__module.roberta.encoder.layer.10.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="1087" name="__module.roberta.encoder.layer.10.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16855,7 +16442,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="807,x.127">
+				<port id="2" precision="FP32" names="715">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -16863,15 +16450,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1227" name="Constant_2647" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="1088" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="808">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1228" name="__module.roberta.encoder.layer.10.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="1089" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -16879,12 +16466,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="809">
+				<port id="2" precision="FP32" names="716,key_layer.21">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -16892,59 +16479,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1229" name="Constant_1059263" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="266274812" size="589824" />
+		<layer id="1090" name="self.roberta.encoder.layer.10.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="265734991" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1230" name="Convert_1059264" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1091" name="Convert_275692" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1231" name="Constant_1059265" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="266864636" size="3072" />
+		<layer id="1092" name="self.roberta.encoder.layer.10.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="266324815" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1093" name="Convert_275695" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1232" name="__module.roberta.encoder.layer.10.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1094" name="self.roberta.encoder.layer.10.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1095" name="self.roberta.encoder.layer.10.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="266325583" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1096" name="self.roberta.encoder.layer.10.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1097" name="self.roberta.encoder.layer.10.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1233" name="__module.roberta.encoder.layer.10.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1098" name="__module.roberta.encoder.layer.10.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16965,8 +16610,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1234" name="Constant_8962" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="266867708" size="3072" />
+		<layer id="1099" name="Constant_9002" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="266327119" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -16975,7 +16620,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1235" name="__module.roberta.encoder.layer.10.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1100" name="__module.roberta.encoder.layer.10.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -16990,22 +16635,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="812,x.129">
+				<port id="2" precision="FP32" names="719">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1236" name="__module.roberta.encoder.layer.10.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="1101" name="Constant_9299" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1237" name="__module.roberta.encoder.layer.10.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="1102" name="__module.roberta.encoder.layer.10.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17018,7 +16663,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="816,x.131">
+				<port id="2" precision="FP32" names="721">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -17026,15 +16671,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1238" name="Constant_2681" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="1103" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="817">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1239" name="__module.roberta.encoder.layer.10.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="1104" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -17042,12 +16687,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="818">
+				<port id="2" precision="FP32" names="722,value_layer.21">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -17055,7 +16700,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1240" name="__module.roberta.encoder.layer.10.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="1105" name="__module.roberta.encoder.layer.10.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17084,7 +16729,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="819,attn_output.41">
+				<port id="4" precision="FP32" names="723,attn_output.41">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -17092,15 +16737,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1241" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="1106" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1242" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="1107" name="__module.roberta.encoder.layer.10.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -17113,7 +16758,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="820,attn_output.43">
+				<port id="2" precision="FP32" names="724,attn_output.43">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -17121,15 +16766,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1243" name="Constant_9179" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="1108" name="Constant_9300" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1244" name="__module.roberta.encoder.layer.10.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="1109" name="__module.roberta.encoder.layer.10.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17143,143 +16788,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="822">
+				<port id="2" precision="FP32" names="726,input.169">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1245" name="__module.roberta.encoder.layer.10.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="266870780" size="3072" />
+		<layer id="1110" name="self.roberta.encoder.layer.10.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="266330191" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1246" name="__module.roberta.encoder.layer.10.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1111" name="Convert_287656" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1247" name="__module.roberta.encoder.layer.10.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="266873852" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1248" name="__module.roberta.encoder.layer.10.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="266873856" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1249" name="__module.roberta.encoder.layer.10.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="266873852" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1250" name="__module.roberta.encoder.layer.10.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="266873856" size="4" />
+		<layer id="1112" name="self.roberta.encoder.layer.10.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="266920015" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1251" name="__module.roberta.encoder.layer.10.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1113" name="Convert_287659" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1252" name="Constant_1059267" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="266873860" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="1114" name="self.roberta.encoder.layer.10.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="1253" name="Convert_1059268" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1254" name="Constant_1059269" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="267463684" size="3072" />
+		<layer id="1115" name="self.roberta.encoder.layer.10.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="266920783" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1255" name="__module.roberta.encoder.layer.10.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1116" name="self.roberta.encoder.layer.10.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1256" name="__module.roberta.encoder.layer.10.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1117" name="self.roberta.encoder.layer.10.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1118" name="__module.roberta.encoder.layer.10.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17300,8 +16926,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1257" name="Constant_8963" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="267466756" size="3072" />
+		<layer id="1119" name="Constant_9003" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="266922319" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -17310,7 +16936,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1258" name="__module.roberta.encoder.layer.10.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1120" name="__module.roberta.encoder.layer.10.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17325,14 +16951,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="827,input.43">
+				<port id="2" precision="FP32" names="731,input.171">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1259" name="__module.roberta.encoder.layer.10.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="1121" name="__module.roberta.encoder.layer.10.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17347,22 +16973,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="829">
+				<port id="2" precision="FP32" names="733,input.173">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1260" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="1122" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1261" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="1123" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17382,8 +17008,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1262" name="Constant_8964" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="267469828" size="3072" />
+		<layer id="1124" name="Constant_9004" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="266925391" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -17392,7 +17018,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1263" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="1125" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17414,8 +17040,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1264" name="Constant_8965" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="267472900" size="3072" />
+		<layer id="1126" name="Constant_9005" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="266928463" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -17424,7 +17050,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1265" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="1127" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17439,143 +17065,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="833,input_tensor.21">
+				<port id="2" precision="FP32" names="737,input.175">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1266" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="267475972" size="3072" />
+		<layer id="1128" name="self.roberta.encoder.layer.10.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="266931535" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1267" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1129" name="Convert_100220" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1268" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="267479044" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1269" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="267479048" size="4" />
+		<layer id="1130" name="self.roberta.encoder.layer.10.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="269290831" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1270" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="267479044" size="4" />
+		<layer id="1131" name="Convert_100223" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1271" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="267479048" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1272" name="__module.roberta.encoder.layer.10.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1132" name="self.roberta.encoder.layer.10.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1273" name="Constant_1059271" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="267479052" size="2359296" />
+		<layer id="1133" name="self.roberta.encoder.layer.10.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="269293903" size="6144" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1274" name="Convert_1059272" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1134" name="self.roberta.encoder.layer.10.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-			</input>
-			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="1275" name="Constant_1059273" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="269838348" size="12288" />
+			</input>
 			<output>
-				<port id="0" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
-					<dim>1</dim>
+					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1276" name="__module.roberta.encoder.layer.10.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1135" name="self.roberta.encoder.layer.10.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>3072</dim>
-					<dim>1</dim>
-				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1277" name="__module.roberta.encoder.layer.10.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1136" name="__module.roberta.encoder.layer.10.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17596,8 +17203,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1278" name="Constant_8966" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="269850636" size="12288" />
+		<layer id="1137" name="Constant_9006" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="269300047" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -17606,7 +17213,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1279" name="__module.roberta.encoder.layer.10.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1138" name="__module.roberta.encoder.layer.10.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17621,14 +17228,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="837">
+				<port id="2" precision="FP32" names="741,input.177">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1280" name="__module.roberta.encoder.layer.10.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="1139" name="__module.roberta.encoder.layer.10.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17638,143 +17245,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="838">
+				<port id="1" precision="FP32" names="742,input.179">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1281" name="__module.roberta.encoder.layer.10.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="269862924" size="12288" />
+		<layer id="1140" name="self.roberta.encoder.layer.10.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="269312335" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1282" name="__module.roberta.encoder.layer.10.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1141" name="Convert_104208" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1283" name="__module.roberta.encoder.layer.10.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="269875212" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1284" name="__module.roberta.encoder.layer.10.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="269875216" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1285" name="__module.roberta.encoder.layer.10.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="269875212" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1286" name="__module.roberta.encoder.layer.10.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="269875216" size="4" />
+		<layer id="1142" name="self.roberta.encoder.layer.10.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="271671631" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1287" name="__module.roberta.encoder.layer.10.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1143" name="Convert_104211" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1288" name="Constant_1059275" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="269875220" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1289" name="Convert_1059276" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1144" name="self.roberta.encoder.layer.10.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1290" name="Constant_1059277" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="272234516" size="3072" />
+		<layer id="1145" name="self.roberta.encoder.layer.10.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="271672399" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1291" name="__module.roberta.encoder.layer.10.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1146" name="self.roberta.encoder.layer.10.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1147" name="self.roberta.encoder.layer.10.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1292" name="__module.roberta.encoder.layer.10.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1148" name="__module.roberta.encoder.layer.10.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17795,8 +17383,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1293" name="Constant_8967" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="272237588" size="3072" />
+		<layer id="1149" name="Constant_9007" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="271673935" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -17805,7 +17393,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1294" name="__module.roberta.encoder.layer.10.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1150" name="__module.roberta.encoder.layer.10.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17820,14 +17408,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="843,input.45">
+				<port id="2" precision="FP32" names="747,input.181">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1295" name="__module.roberta.encoder.layer.10.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="1151" name="__module.roberta.encoder.layer.10.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17842,22 +17430,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="845">
+				<port id="2" precision="FP32" names="749,input.183">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1296" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="1152" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1297" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="1153" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17877,8 +17465,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1298" name="Constant_8968" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="272240660" size="3072" />
+		<layer id="1154" name="Constant_9008" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="271677007" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -17887,7 +17475,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1299" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="1155" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17909,8 +17497,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1300" name="Constant_8969" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="272243732" size="3072" />
+		<layer id="1156" name="Constant_9009" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="271680079" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -17919,7 +17507,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1301" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="1157" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -17934,143 +17522,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="849,hidden_states.67">
+				<port id="2" precision="FP32" names="753,hidden_states.67">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1302" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="272246804" size="3072" />
+		<layer id="1158" name="self.roberta.encoder.layer.11.attention.self.query.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="271683151" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1303" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1159" name="Convert_299620" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1304" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="272249876" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1305" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="272249880" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1306" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="272249876" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1307" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="272249880" size="4" />
+		<layer id="1160" name="self.roberta.encoder.layer.11.attention.self.query.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="272272975" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1308" name="__module.roberta.encoder.layer.10.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1161" name="Convert_299623" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1309" name="Constant_1059279" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="272249884" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="1162" name="self.roberta.encoder.layer.11.attention.self.query.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="1310" name="Convert_1059280" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1311" name="Constant_1059281" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="272839708" size="3072" />
+		<layer id="1163" name="self.roberta.encoder.layer.11.attention.self.query.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="272273743" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1312" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1164" name="self.roberta.encoder.layer.11.attention.self.query.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1165" name="self.roberta.encoder.layer.11.attention.self.query.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1313" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1166" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18091,8 +17660,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1314" name="Constant_8970" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="272842780" size="3072" />
+		<layer id="1167" name="Constant_9010" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="272275279" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -18101,7 +17670,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1315" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1168" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18116,67 +17685,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="862,x.133">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1316" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="272845852" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1317" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="272845856" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1318" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="272845852" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1319" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="272845856" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1320" name="__module.roberta.encoder.layer.11.attention.self.query/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
-			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
-			</input>
-			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="766">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1321" name="__module.roberta.encoder.layer.11.attention.self/prim::ListConstruct" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="1169" name="Constant_9301" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1322" name="__module.roberta.encoder.layer.11.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
+		<layer id="1170" name="__module.roberta.encoder.layer.11.attention.self/aten::view/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18189,7 +17713,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="866,x.135">
+				<port id="2" precision="FP32" names="768">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -18197,15 +17721,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1323" name="Constant_2847" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="1171" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Constant" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="867">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1324" name="__module.roberta.encoder.layer.11.attention.self/aten::permute/Transpose" type="Transpose" version="opset1">
+		<layer id="1172" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -18213,12 +17737,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="868">
+				<port id="2" precision="FP32" names="769,query_layer">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -18226,165 +17750,178 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1325" name="Constant_1059283" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="272845860" size="589824" />
+		<layer id="1173" name="self.roberta.encoder.layer.11.attention.self.key.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="272278351" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1326" name="Convert_1059284" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1174" name="Convert_295632" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1327" name="Constant_1059285" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="273435684" size="3072" />
+		<layer id="1175" name="self.roberta.encoder.layer.11.attention.self.key.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="272868175" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1328" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1176" name="Convert_295635" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>768</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1329" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
-			<data transpose_a="false" transpose_b="true" />
+		<layer id="1177" name="self.roberta.encoder.layer.11.attention.self.key.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
 					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1330" name="Constant_8971" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="273438756" size="3072" />
+		<layer id="1178" name="self.roberta.encoder.layer.11.attention.self.key.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="272868943" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1331" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1179" name="self.roberta.encoder.layer.11.attention.self.key.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="871,x.137">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1332" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/Add/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="273441828" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1333" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/Add/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="273441832" size="4" />
+		<layer id="1180" name="self.roberta.encoder.layer.11.attention.self.key.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1334" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/Add/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="273441828" size="4" />
+		<layer id="1181" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/MatMul" type="MatMul" version="opset1">
+			<data transpose_a="false" transpose_b="true" />
+			<input>
+				<port id="0" precision="FP32">
+					<dim>-1</dim>
+					<dim>-1</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="2" precision="FP32">
+					<dim>-1</dim>
+					<dim>-1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1335" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/Add/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="273441832" size="4" />
+		<layer id="1182" name="Constant_9011" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="272870479" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1336" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/Add/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1183" name="__module.roberta.encoder.layer.11.attention.self.key/aten::linear/Add" type="Add" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
+				<port id="1" precision="FP32">
+					<dim>1</dim>
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
 			</input>
 			<output>
-				<port id="5" precision="FP32">
+				<port id="2" precision="FP32" names="772">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1337" name="__module.roberta.encoder.layer.11.attention.self/prim::ListConstruct_1" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="1184" name="Constant_9302" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1338" name="__module.roberta.encoder.layer.11.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
+		<layer id="1185" name="__module.roberta.encoder.layer.11.attention.self/aten::view/Reshape_1" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18397,7 +17934,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="875,x.139">
+				<port id="2" precision="FP32" names="774">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -18405,15 +17942,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1339" name="Constant_2881" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="1186" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Constant_1" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="876">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1340" name="__module.roberta.encoder.layer.11.attention.self/aten::permute/Transpose_1" type="Transpose" version="opset1">
+		<layer id="1187" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Transpose_1" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -18421,12 +17958,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="877">
+				<port id="2" precision="FP32" names="775,key_layer">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -18434,59 +17971,117 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1341" name="Constant_1059287" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="273441836" size="589824" />
+		<layer id="1188" name="self.roberta.encoder.layer.11.attention.self.value.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="272873551" size="589824" />
 			<output>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1342" name="Convert_1059288" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1189" name="Convert_291644" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="U8">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1343" name="Constant_1059289" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="274031660" size="3072" />
+		<layer id="1190" name="self.roberta.encoder.layer.11.attention.self.value.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="273463375" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1191" name="Convert_291647" type="Convert" version="opset1">
+			<data destination_type="f16" />
+			<input>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1344" name="__module.roberta.encoder.layer.11.attention.self.value/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1192" name="self.roberta.encoder.layer.11.attention.self.value.weight/zero_point/subtract" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1345" name="__module.roberta.encoder.layer.11.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1193" name="self.roberta.encoder.layer.11.attention.self.value.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="273464143" size="1536" />
+			<output>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1194" name="self.roberta.encoder.layer.11.attention.self.value.weight/fq_weights_1" type="Multiply" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
+			</input>
+			<output>
+				<port id="2" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1195" name="self.roberta.encoder.layer.11.attention.self.value.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1196" name="__module.roberta.encoder.layer.11.attention.self.value/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18507,8 +18102,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1346" name="Constant_8972" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="274034732" size="3072" />
+		<layer id="1197" name="Constant_9012" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="273465679" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -18517,7 +18112,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1347" name="__module.roberta.encoder.layer.11.attention.self.value/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1198" name="__module.roberta.encoder.layer.11.attention.self.value/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18532,22 +18127,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="880,x.141">
+				<port id="2" precision="FP32" names="778">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1348" name="__module.roberta.encoder.layer.11.attention.self/prim::ListConstruct_2" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008460" size="32" />
+		<layer id="1199" name="Constant_9303" type="Const" version="opset1">
+			<data element_type="i64" shape="4" offset="193754087" size="32" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1349" name="__module.roberta.encoder.layer.11.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
+		<layer id="1200" name="__module.roberta.encoder.layer.11.attention.self/aten::view/Reshape_2" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18560,7 +18155,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="884,x">
+				<port id="2" precision="FP32" names="780">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -18568,15 +18163,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1350" name="Constant_2915" type="Const" version="opset1">
-			<data element_type="i64" shape="4" offset="194008492" size="32" />
+		<layer id="1201" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Constant_2" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
-				<port id="0" precision="I64" names="885">
+				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1351" name="__module.roberta.encoder.layer.11.attention.self/aten::permute/Transpose_2" type="Transpose" version="opset1">
+		<layer id="1202" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Transpose_2" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -18584,12 +18179,12 @@
 					<dim>12</dim>
 					<dim>64</dim>
 				</port>
-				<port id="1" precision="I64">
+				<port id="1" precision="I32">
 					<dim>4</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="886">
+				<port id="2" precision="FP32" names="781,value_layer">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -18597,7 +18192,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1352" name="__module.roberta.encoder.layer.11.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
+		<layer id="1203" name="__module.roberta.encoder.layer.11.attention.self/aten::scaled_dot_product_attention/ScaledDotProductAttention" type="ScaledDotProductAttention" version="opset13">
 			<data causal="false" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18626,7 +18221,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="4" precision="FP32" names="887,attn_output.45">
+				<port id="4" precision="FP32" names="782,attn_output.45">
 					<dim>-1</dim>
 					<dim>12</dim>
 					<dim>-1</dim>
@@ -18634,15 +18229,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1353" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Constant" type="Const" version="opset1">
-			<data element_type="i32" shape="4" offset="195200500" size="16" />
+		<layer id="1204" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Constant_3" type="Const" version="opset1">
+			<data element_type="i32" shape="4" offset="193754119" size="16" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>4</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1354" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Transpose" type="Transpose" version="opset1">
+		<layer id="1205" name="__module.roberta.encoder.layer.11.attention.self/aten::transpose/Transpose_3" type="Transpose" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -18655,7 +18250,7 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="888,attn_output">
+				<port id="2" precision="FP32" names="783,attn_output">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>12</dim>
@@ -18663,15 +18258,15 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1355" name="Constant_9180" type="Const" version="opset1">
-			<data element_type="i64" shape="3" offset="195200516" size="24" />
+		<layer id="1206" name="Constant_9304" type="Const" version="opset1">
+			<data element_type="i64" shape="3" offset="194944567" size="24" />
 			<output>
 				<port id="0" precision="I64">
 					<dim>3</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1356" name="__module.roberta.encoder.layer.11.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
+		<layer id="1207" name="__module.roberta.encoder.layer.11.attention.self/aten::reshape/Reshape" type="Reshape" version="opset1">
 			<data special_zero="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18685,143 +18280,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="890">
+				<port id="2" precision="FP32" names="785,input.185">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1357" name="__module.roberta.encoder.layer.11.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="274037804" size="3072" />
+		<layer id="1208" name="self.roberta.encoder.layer.11.attention.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="273468751" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1358" name="__module.roberta.encoder.layer.11.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1209" name="Convert_303608" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1359" name="__module.roberta.encoder.layer.11.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="274040876" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1360" name="__module.roberta.encoder.layer.11.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="274040880" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1361" name="__module.roberta.encoder.layer.11.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="274040876" size="4" />
+		<layer id="1210" name="self.roberta.encoder.layer.11.attention.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="274058575" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1362" name="__module.roberta.encoder.layer.11.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="274040880" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1363" name="__module.roberta.encoder.layer.11.attention.self/aten::reshape/Reshape_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1211" name="Convert_303611" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1364" name="Constant_1059291" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="274040884" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="1212" name="self.roberta.encoder.layer.11.attention.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="1365" name="Convert_1059292" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1366" name="Constant_1059293" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="274630708" size="3072" />
+		<layer id="1213" name="self.roberta.encoder.layer.11.attention.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="274059343" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1367" name="__module.roberta.encoder.layer.11.attention.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1214" name="self.roberta.encoder.layer.11.attention.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1368" name="__module.roberta.encoder.layer.11.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1215" name="self.roberta.encoder.layer.11.attention.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1216" name="__module.roberta.encoder.layer.11.attention.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18842,8 +18418,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1369" name="Constant_8973" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="274633780" size="3072" />
+		<layer id="1217" name="Constant_9013" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="274060879" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -18852,7 +18428,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1370" name="__module.roberta.encoder.layer.11.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1218" name="__module.roberta.encoder.layer.11.attention.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18867,14 +18443,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="895,input.47">
+				<port id="2" precision="FP32" names="790,input.187">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1371" name="__module.roberta.encoder.layer.11.attention.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="1219" name="__module.roberta.encoder.layer.11.attention.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18889,22 +18465,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="897">
+				<port id="2" precision="FP32" names="792,input.189">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1372" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="1220" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1373" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="1221" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18924,8 +18500,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1374" name="Constant_8974" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="274636852" size="3072" />
+		<layer id="1222" name="Constant_9014" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="274063951" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -18934,7 +18510,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1375" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="1223" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18956,8 +18532,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1376" name="Constant_8975" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="274639924" size="3072" />
+		<layer id="1224" name="Constant_9015" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="274067023" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -18966,7 +18542,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1377" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="1225" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -18981,143 +18557,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="901,input_tensor">
+				<port id="2" precision="FP32" names="796,input.191">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1378" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="274642996" size="3072" />
+		<layer id="1226" name="self.roberta.encoder.layer.11.intermediate.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 768" offset="274070095" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1379" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1227" name="Convert_108196" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1380" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="274646068" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1381" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="274646072" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1382" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="274646068" size="4" />
+		<layer id="1228" name="self.roberta.encoder.layer.11.intermediate.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="3072, 1" offset="276429391" size="3072" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1383" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="274646072" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1384" name="__module.roberta.encoder.layer.11.attention.output.LayerNorm/aten::layer_norm/Add_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1229" name="Convert_108199" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1385" name="Constant_1059295" type="Const" version="opset1">
-			<data element_type="i8" shape="3072, 768" offset="274646076" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
-					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1386" name="Convert_1059296" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1230" name="self.roberta.encoder.layer.11.intermediate.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>3072</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1387" name="Constant_1059297" type="Const" version="opset1">
-			<data element_type="f32" shape="3072, 1" offset="277005372" size="12288" />
+		<layer id="1231" name="self.roberta.encoder.layer.11.intermediate.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="3072, 1" offset="276432463" size="6144" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1388" name="__module.roberta.encoder.layer.11.intermediate.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1232" name="self.roberta.encoder.layer.11.intermediate.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>3072</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1233" name="self.roberta.encoder.layer.11.intermediate.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>3072</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>3072</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1389" name="__module.roberta.encoder.layer.11.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1234" name="__module.roberta.encoder.layer.11.intermediate.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19138,8 +18695,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1390" name="Constant_8976" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="277017660" size="12288" />
+		<layer id="1235" name="Constant_9016" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 3072" offset="276438607" size="12288" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -19148,7 +18705,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1391" name="__module.roberta.encoder.layer.11.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1236" name="__module.roberta.encoder.layer.11.intermediate.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19163,14 +18720,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="905">
+				<port id="2" precision="FP32" names="800,input.193">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1392" name="__module.roberta.encoder.layer.11.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
+		<layer id="1237" name="__module.roberta.encoder.layer.11.intermediate.intermediate_act_fn/aten::gelu/Gelu" type="Gelu" version="opset7">
 			<data approximation_mode="ERF" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19180,143 +18737,124 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="906">
+				<port id="1" precision="FP32" names="801,input.195">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1393" name="__module.roberta.encoder.layer.11.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 3072" offset="277029948" size="12288" />
+		<layer id="1238" name="self.roberta.encoder.layer.11.output.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 3072" offset="276450895" size="2359296" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1394" name="__module.roberta.encoder.layer.11.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1239" name="Convert_112184" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1395" name="__module.roberta.encoder.layer.11.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="277042236" size="4" />
+		<layer id="1240" name="self.roberta.encoder.layer.11.output.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="278810191" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1396" name="__module.roberta.encoder.layer.11.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="277042240" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1397" name="__module.roberta.encoder.layer.11.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="277042236" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1398" name="__module.roberta.encoder.layer.11.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="277042240" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1399" name="__module.roberta.encoder.layer.11.intermediate.intermediate_act_fn/aten::gelu/Gelu_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1241" name="Convert_112187" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>-1</dim>
-					<dim>3072</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1400" name="Constant_1059299" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 3072" offset="277042244" size="2359296" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
-					<dim>3072</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1401" name="Convert_1059300" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1242" name="self.roberta.encoder.layer.11.output.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1402" name="Constant_1059301" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="279401540" size="3072" />
+		<layer id="1243" name="self.roberta.encoder.layer.11.output.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="278810959" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1403" name="__module.roberta.encoder.layer.11.output.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1244" name="self.roberta.encoder.layer.11.output.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>3072</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1404" name="__module.roberta.encoder.layer.11.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1245" name="self.roberta.encoder.layer.11.output.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>3072</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1246" name="__module.roberta.encoder.layer.11.output.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19337,8 +18875,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1405" name="Constant_8977" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="279404612" size="3072" />
+		<layer id="1247" name="Constant_9017" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="278812495" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -19347,7 +18885,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1406" name="__module.roberta.encoder.layer.11.output.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1248" name="__module.roberta.encoder.layer.11.output.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19362,14 +18900,14 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="911,input.49">
+				<port id="2" precision="FP32" names="806,input.197">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1407" name="__module.roberta.encoder.layer.11.output/aten::add/Add" type="Add" version="opset1">
+		<layer id="1249" name="__module.roberta.encoder.layer.11.output/aten::add/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19384,22 +18922,22 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="913">
+				<port id="2" precision="FP32" names="808,input.199">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1408" name="__module.roberta.encoder.layer.11.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
-			<data element_type="i32" shape="1" offset="193403256" size="4" />
+		<layer id="1250" name="__module.roberta.encoder.layer.11.output.LayerNorm/aten::layer_norm/Multiply" type="Const" version="opset1">
+			<data element_type="i32" shape="1" offset="193152739" size="4" />
 			<output>
 				<port id="0" precision="I32">
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1409" name="__module.roberta.encoder.layer.11.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
+		<layer id="1251" name="__module.roberta.encoder.layer.11.output.LayerNorm/aten::layer_norm/MVN" type="MVN" version="opset6">
 			<data eps="9.9999997473787516e-06" normalize_variance="true" eps_mode="INSIDE_SQRT" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19419,8 +18957,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1410" name="Constant_8978" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="279407684" size="3072" />
+		<layer id="1252" name="Constant_9018" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="278815567" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -19429,7 +18967,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1411" name="__module.roberta.encoder.layer.11.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
+		<layer id="1253" name="__module.roberta.encoder.layer.11.output.LayerNorm/aten::layer_norm/Multiply_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19451,8 +18989,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1412" name="Constant_8979" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1, 768" offset="279410756" size="3072" />
+		<layer id="1254" name="Constant_9019" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1, 768" offset="278818639" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -19461,7 +18999,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1413" name="__module.roberta.encoder.layer.11.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
+		<layer id="1255" name="__module.roberta.encoder.layer.11.output.LayerNorm/aten::layer_norm/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19476,26 +19014,26 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="917,926,features">
+				<port id="2" precision="FP32" names="812,821,features">
 					<dim>-1</dim>
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1414" name="922" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="193006432" size="8" />
+		<layer id="1256" name="817" type="Const" version="opset1">
+			<data element_type="i64" shape="" offset="192756429" size="8" />
 			<output>
-				<port id="0" precision="I64" names="922" />
+				<port id="0" precision="I64" names="817" />
 			</output>
 		</layer>
-		<layer id="1415" name="920" type="Const" version="opset1">
-			<data element_type="i64" shape="" offset="193006440" size="8" />
+		<layer id="1257" name="815" type="Const" version="opset1">
+			<data element_type="i64" shape="" offset="192756437" size="8" />
 			<output>
-				<port id="0" precision="I64" names="920" />
+				<port id="0" precision="I64" names="815" />
 			</output>
 		</layer>
-		<layer id="1416" name="__module.classifier/aten::select/Gather" type="Gather" version="opset8">
+		<layer id="1258" name="__module.classifier/aten::select/Gather" type="Gather" version="opset8">
 			<data batch_dims="0" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19507,136 +19045,123 @@
 				<port id="2" precision="I64" />
 			</input>
 			<output>
-				<port id="3" precision="FP32" names="927,928,input.51">
+				<port id="3" precision="FP32" names="822,823,input.201">
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1417" name="__module.classifier/aten::select/Gather_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 768" offset="279413828" size="3072" />
+		<layer id="1259" name="self.classifier.dense.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 768" offset="278821711" size="589824" />
 			<output>
-				<port id="0" precision="FP32">
-					<dim>1</dim>
+				<port id="0" precision="U8">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1418" name="__module.classifier/aten::select/Gather_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1260" name="Convert_307596" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
-					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1419" name="__module.classifier/aten::select/Gather_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="279416900" size="4" />
+		<layer id="1261" name="self.classifier.dense.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="768, 1" offset="279411535" size="768" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1420" name="__module.classifier/aten::select/Gather_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="279416904" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1421" name="__module.classifier/aten::select/Gather_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="279416900" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1422" name="__module.classifier/aten::select/Gather_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="279416904" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>768</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1423" name="__module.classifier/aten::select/Gather_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1262" name="Convert_307599" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
+				<port id="0" precision="U8">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1424" name="Constant_1059303" type="Const" version="opset1">
-			<data element_type="i8" shape="768, 768" offset="279416908" size="589824" />
-			<output>
-				<port id="0" precision="I8">
+		<layer id="1263" name="self.classifier.dense.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
+			<input>
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-			</output>
-		</layer>
-		<layer id="1425" name="Convert_1059304" type="Convert" version="opset1">
-			<data destination_type="f32" />
-			<input>
-				<port id="0" precision="I8">
-					<dim>768</dim>
+				<port id="1" precision="FP16">
 					<dim>768</dim>
+					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1426" name="Constant_1059305" type="Const" version="opset1">
-			<data element_type="f32" shape="768, 1" offset="280006732" size="3072" />
+		<layer id="1264" name="self.classifier.dense.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="768, 1" offset="279412303" size="1536" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1427" name="__module.classifier.dense/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1265" name="self.classifier.dense.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>768</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>768</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1428" name="__module.classifier.dense/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1266" name="self.classifier.dense.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
+					<dim>768</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1267" name="__module.classifier.dense/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19655,8 +19180,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1429" name="Constant_8980" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 768" offset="280009804" size="3072" />
+		<layer id="1268" name="Constant_9020" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 768" offset="279413839" size="3072" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -19664,7 +19189,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1430" name="__module.classifier.dense/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1269" name="__module.classifier.dense/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19677,13 +19202,13 @@
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32" names="932">
+				<port id="2" precision="FP32" names="827,x">
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1431" name="__module.classifier/aten::tanh/Tanh" type="Tanh" version="opset1">
+		<layer id="1270" name="__module.classifier/aten::tanh/Tanh" type="Tanh" version="opset1">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -19691,136 +19216,123 @@
 				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32" names="933,input">
+				<port id="1" precision="FP32" names="828,input.205">
 					<dim>-1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1432" name="__module.classifier/aten::tanh/Tanh_0_0/nncf_smooth_quant/scale" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 768" offset="280012876" size="3072" />
+		<layer id="1271" name="self.classifier.out_proj.weight" type="Const" version="opset1">
+			<data element_type="u8" shape="1, 768" offset="279416911" size="768" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1433" name="__module.classifier/aten::tanh/Tanh_0_0/nncf_smooth_quant" type="Multiply" version="opset1">
-			<data auto_broadcast="numpy" />
+		<layer id="1272" name="Convert_311606" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-				<port id="1" precision="FP32">
+				<port id="0" precision="U8">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
-					<dim>-1</dim>
+				<port id="1" precision="FP16">
+					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1434" name="__module.classifier/aten::tanh/Tanh_0_0/nncf_smooth_quant/fq_output_0/input_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="280015948" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1435" name="__module.classifier/aten::tanh/Tanh_0_0/nncf_smooth_quant/fq_output_0/input_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="280015952" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1436" name="__module.classifier/aten::tanh/Tanh_0_0/nncf_smooth_quant/fq_output_0/output_low" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="280015948" size="4" />
+		<layer id="1273" name="self.classifier.out_proj.weight/zero_point" type="Const" version="opset1">
+			<data element_type="u8" shape="1, 1" offset="279417679" size="1" />
 			<output>
-				<port id="0" precision="FP32" />
-			</output>
-		</layer>
-		<layer id="1437" name="__module.classifier/aten::tanh/Tanh_0_0/nncf_smooth_quant/fq_output_0/output_high" type="Const" version="opset1">
-			<data element_type="f32" shape="" offset="280015952" size="4" />
-			<output>
-				<port id="0" precision="FP32" />
+				<port id="0" precision="U8">
+					<dim>1</dim>
+					<dim>1</dim>
+				</port>
 			</output>
 		</layer>
-		<layer id="1438" name="__module.classifier/aten::tanh/Tanh_0_0/nncf_smooth_quant/fq_output_0" type="FakeQuantize" version="opset1">
-			<data levels="256" auto_broadcast="numpy" />
+		<layer id="1274" name="Convert_311609" type="Convert" version="opset1">
+			<data destination_type="f16" />
 			<input>
-				<port id="0" precision="FP32">
-					<dim>-1</dim>
-					<dim>768</dim>
+				<port id="0" precision="U8">
+					<dim>1</dim>
+					<dim>1</dim>
 				</port>
-				<port id="1" precision="FP32" />
-				<port id="2" precision="FP32" />
-				<port id="3" precision="FP32" />
-				<port id="4" precision="FP32" />
 			</input>
 			<output>
-				<port id="5" precision="FP32">
-					<dim>-1</dim>
-					<dim>768</dim>
-				</port>
-			</output>
-		</layer>
-		<layer id="1439" name="Constant_1059307" type="Const" version="opset1">
-			<data element_type="i8" shape="1, 768" offset="280015956" size="768" />
-			<output>
-				<port id="0" precision="I8">
+				<port id="1" precision="FP16">
+					<dim>1</dim>
 					<dim>1</dim>
-					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1440" name="Convert_1059308" type="Convert" version="opset1">
-			<data destination_type="f32" />
+		<layer id="1275" name="self.classifier.out_proj.weight/zero_point/subtract" type="Subtract" version="opset1">
+			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="I8">
+				<port id="0" precision="FP16">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
+				<port id="1" precision="FP16">
+					<dim>1</dim>
+					<dim>1</dim>
+				</port>
 			</input>
 			<output>
-				<port id="1" precision="FP32">
+				<port id="2" precision="FP16">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1441" name="Constant_1059309" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1" offset="280016724" size="4" />
+		<layer id="1276" name="self.classifier.out_proj.weight/scale" type="Const" version="opset1">
+			<data element_type="f16" shape="1, 1" offset="279417680" size="2" />
 			<output>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1442" name="__module.classifier.out_proj/aten::linear/MatMul/fq_weights_1" type="Multiply" version="opset1">
+		<layer id="1277" name="self.classifier.out_proj.weight/fq_weights_1" type="Multiply" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
-				<port id="0" precision="FP32">
+				<port id="0" precision="FP16">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
-				<port id="1" precision="FP32">
+				<port id="1" precision="FP16">
 					<dim>1</dim>
 					<dim>1</dim>
 				</port>
 			</input>
 			<output>
-				<port id="2" precision="FP32">
+				<port id="2" precision="FP16">
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
+			</output>
+		</layer>
+		<layer id="1278" name="self.classifier.out_proj.weight/fq_weights_1/convert" type="Convert" version="opset1">
+			<data destination_type="f32" />
+			<input>
+				<port id="0" precision="FP16">
+					<dim>1</dim>
+					<dim>768</dim>
+				</port>
+			</input>
+			<output>
+				<port id="1" precision="FP32">
 					<dim>1</dim>
 					<dim>768</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1443" name="__module.classifier.out_proj/aten::linear/MatMul" type="MatMul" version="opset1">
+		<layer id="1279" name="__module.classifier.out_proj/aten::linear/MatMul" type="MatMul" version="opset1">
 			<data transpose_a="false" transpose_b="true" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19839,8 +19351,8 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1444" name="Constant_8981" type="Const" version="opset1">
-			<data element_type="f32" shape="1, 1" offset="280016728" size="4" />
+		<layer id="1280" name="Constant_9021" type="Const" version="opset1">
+			<data element_type="f32" shape="1, 1" offset="279417682" size="4" />
 			<output>
 				<port id="0" precision="FP32">
 					<dim>1</dim>
@@ -19848,7 +19360,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1445" name="__module.classifier.out_proj/aten::linear/Add" type="Add" version="opset1">
+		<layer id="1281" name="__module.classifier.out_proj/aten::linear/Add" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="FP32">
@@ -19867,7 +19379,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="1446" name="Result_3145" type="Result" version="opset1" output_names="logits">
+		<layer id="1282" name="Result_3719" type="Result" version="opset1" output_names="logits">
 			<input>
 				<port id="0" precision="FP32">
 					<dim>-1</dim>
@@ -19877,1543 +19389,1387 @@
 		</layer>
 	</layers>
 	<edges>
-		<edge from-layer="0" from-port="0" to-layer="101" to-port="0" />
-		<edge from-layer="1" from-port="0" to-layer="6" to-port="0" />
-		<edge from-layer="1" from-port="0" to-layer="32" to-port="0" />
-		<edge from-layer="1" from-port="0" to-layer="15" to-port="0" />
+		<edge from-layer="0" from-port="0" to-layer="108" to-port="0" />
+		<edge from-layer="1" from-port="0" to-layer="44" to-port="0" />
+		<edge from-layer="1" from-port="0" to-layer="23" to-port="0" />
+		<edge from-layer="1" from-port="0" to-layer="10" to-port="0" />
 		<edge from-layer="2" from-port="0" to-layer="3" to-port="0" />
-		<edge from-layer="3" from-port="1" to-layer="5" to-port="0" />
-		<edge from-layer="4" from-port="0" to-layer="5" to-port="1" />
-		<edge from-layer="5" from-port="2" to-layer="8" to-port="0" />
-		<edge from-layer="6" from-port="1" to-layer="8" to-port="1" />
-		<edge from-layer="7" from-port="0" to-layer="8" to-port="2" />
-		<edge from-layer="8" from-port="3" to-layer="26" to-port="0" />
-		<edge from-layer="9" from-port="0" to-layer="10" to-port="0" />
-		<edge from-layer="10" from-port="1" to-layer="12" to-port="0" />
-		<edge from-layer="11" from-port="0" to-layer="12" to-port="1" />
-		<edge from-layer="12" from-port="2" to-layer="25" to-port="0" />
-		<edge from-layer="13" from-port="0" to-layer="21" to-port="0" />
-		<edge from-layer="14" from-port="0" to-layer="21" to-port="1" />
-		<edge from-layer="15" from-port="1" to-layer="22" to-port="1" />
-		<edge from-layer="15" from-port="1" to-layer="18" to-port="0" />
-		<edge from-layer="15" from-port="1" to-layer="110" to-port="0" />
-		<edge from-layer="15" from-port="1" to-layer="106" to-port="0" />
-		<edge from-layer="16" from-port="0" to-layer="18" to-port="1" />
-		<edge from-layer="17" from-port="0" to-layer="18" to-port="2" />
-		<edge from-layer="18" from-port="3" to-layer="21" to-port="2" />
-		<edge from-layer="19" from-port="0" to-layer="21" to-port="3" />
-		<edge from-layer="20" from-port="0" to-layer="21" to-port="4" />
-		<edge from-layer="21" from-port="5" to-layer="22" to-port="0" />
-		<edge from-layer="22" from-port="2" to-layer="23" to-port="0" />
-		<edge from-layer="23" from-port="1" to-layer="25" to-port="1" />
-		<edge from-layer="24" from-port="0" to-layer="25" to-port="2" />
-		<edge from-layer="25" from-port="3" to-layer="26" to-port="1" />
-		<edge from-layer="26" from-port="2" to-layer="43" to-port="0" />
-		<edge from-layer="27" from-port="0" to-layer="28" to-port="0" />
-		<edge from-layer="28" from-port="1" to-layer="30" to-port="0" />
-		<edge from-layer="29" from-port="0" to-layer="30" to-port="1" />
-		<edge from-layer="30" from-port="2" to-layer="42" to-port="0" />
-		<edge from-layer="31" from-port="0" to-layer="32" to-port="1" />
-		<edge from-layer="32" from-port="2" to-layer="33" to-port="0" />
-		<edge from-layer="33" from-port="1" to-layer="36" to-port="1" />
-		<edge from-layer="33" from-port="1" to-layer="35" to-port="0" />
-		<edge from-layer="34" from-port="0" to-layer="101" to-port="1" />
-		<edge from-layer="34" from-port="0" to-layer="35" to-port="1" />
-		<edge from-layer="35" from-port="2" to-layer="36" to-port="0" />
-		<edge from-layer="36" from-port="2" to-layer="37" to-port="0" />
-		<edge from-layer="37" from-port="1" to-layer="39" to-port="0" />
-		<edge from-layer="38" from-port="0" to-layer="39" to-port="1" />
-		<edge from-layer="39" from-port="2" to-layer="40" to-port="0" />
-		<edge from-layer="40" from-port="1" to-layer="42" to-port="1" />
-		<edge from-layer="41" from-port="0" to-layer="42" to-port="2" />
-		<edge from-layer="42" from-port="3" to-layer="43" to-port="1" />
-		<edge from-layer="43" from-port="2" to-layer="45" to-port="0" />
-		<edge from-layer="44" from-port="0" to-layer="45" to-port="1" />
-		<edge from-layer="45" from-port="2" to-layer="47" to-port="0" />
+		<edge from-layer="3" from-port="1" to-layer="6" to-port="0" />
+		<edge from-layer="4" from-port="0" to-layer="5" to-port="0" />
+		<edge from-layer="5" from-port="1" to-layer="6" to-port="1" />
+		<edge from-layer="6" from-port="2" to-layer="8" to-port="0" />
+		<edge from-layer="7" from-port="0" to-layer="8" to-port="1" />
+		<edge from-layer="8" from-port="2" to-layer="9" to-port="0" />
+		<edge from-layer="9" from-port="1" to-layer="12" to-port="0" />
+		<edge from-layer="10" from-port="1" to-layer="12" to-port="1" />
+		<edge from-layer="11" from-port="0" to-layer="12" to-port="2" />
+		<edge from-layer="12" from-port="3" to-layer="34" to-port="0" />
+		<edge from-layer="13" from-port="0" to-layer="14" to-port="0" />
+		<edge from-layer="14" from-port="1" to-layer="17" to-port="0" />
+		<edge from-layer="15" from-port="0" to-layer="16" to-port="0" />
+		<edge from-layer="16" from-port="1" to-layer="17" to-port="1" />
+		<edge from-layer="17" from-port="2" to-layer="19" to-port="0" />
+		<edge from-layer="18" from-port="0" to-layer="19" to-port="1" />
+		<edge from-layer="19" from-port="2" to-layer="20" to-port="0" />
+		<edge from-layer="20" from-port="1" to-layer="33" to-port="0" />
+		<edge from-layer="21" from-port="0" to-layer="29" to-port="0" />
+		<edge from-layer="22" from-port="0" to-layer="29" to-port="1" />
+		<edge from-layer="23" from-port="1" to-layer="26" to-port="0" />
+		<edge from-layer="23" from-port="1" to-layer="30" to-port="1" />
+		<edge from-layer="23" from-port="1" to-layer="117" to-port="0" />
+		<edge from-layer="23" from-port="1" to-layer="113" to-port="0" />
+		<edge from-layer="24" from-port="0" to-layer="26" to-port="1" />
+		<edge from-layer="25" from-port="0" to-layer="26" to-port="2" />
+		<edge from-layer="26" from-port="3" to-layer="29" to-port="2" />
+		<edge from-layer="27" from-port="0" to-layer="29" to-port="3" />
+		<edge from-layer="28" from-port="0" to-layer="29" to-port="4" />
+		<edge from-layer="29" from-port="5" to-layer="30" to-port="0" />
+		<edge from-layer="30" from-port="2" to-layer="31" to-port="0" />
+		<edge from-layer="31" from-port="1" to-layer="33" to-port="1" />
+		<edge from-layer="32" from-port="0" to-layer="33" to-port="2" />
+		<edge from-layer="33" from-port="3" to-layer="34" to-port="1" />
+		<edge from-layer="34" from-port="2" to-layer="55" to-port="0" />
+		<edge from-layer="35" from-port="0" to-layer="36" to-port="0" />
+		<edge from-layer="36" from-port="1" to-layer="39" to-port="0" />
+		<edge from-layer="37" from-port="0" to-layer="38" to-port="0" />
+		<edge from-layer="38" from-port="1" to-layer="39" to-port="1" />
+		<edge from-layer="39" from-port="2" to-layer="41" to-port="0" />
+		<edge from-layer="40" from-port="0" to-layer="41" to-port="1" />
+		<edge from-layer="41" from-port="2" to-layer="42" to-port="0" />
+		<edge from-layer="42" from-port="1" to-layer="54" to-port="0" />
+		<edge from-layer="43" from-port="0" to-layer="44" to-port="1" />
+		<edge from-layer="44" from-port="2" to-layer="45" to-port="0" />
+		<edge from-layer="45" from-port="1" to-layer="48" to-port="1" />
+		<edge from-layer="45" from-port="1" to-layer="47" to-port="0" />
 		<edge from-layer="46" from-port="0" to-layer="47" to-port="1" />
-		<edge from-layer="47" from-port="2" to-layer="49" to-port="0" />
-		<edge from-layer="48" from-port="0" to-layer="49" to-port="1" />
-		<edge from-layer="49" from-port="2" to-layer="139" to-port="1" />
-		<edge from-layer="49" from-port="2" to-layer="51" to-port="0" />
+		<edge from-layer="46" from-port="0" to-layer="108" to-port="1" />
+		<edge from-layer="47" from-port="2" to-layer="48" to-port="0" />
+		<edge from-layer="48" from-port="2" to-layer="49" to-port="0" />
+		<edge from-layer="49" from-port="1" to-layer="51" to-port="0" />
 		<edge from-layer="50" from-port="0" to-layer="51" to-port="1" />
-		<edge from-layer="51" from-port="2" to-layer="56" to-port="0" />
-		<edge from-layer="52" from-port="0" to-layer="56" to-port="1" />
-		<edge from-layer="53" from-port="0" to-layer="56" to-port="2" />
-		<edge from-layer="54" from-port="0" to-layer="56" to-port="3" />
-		<edge from-layer="55" from-port="0" to-layer="56" to-port="4" />
-		<edge from-layer="56" from-port="5" to-layer="77" to-port="0" />
-		<edge from-layer="56" from-port="5" to-layer="93" to-port="0" />
-		<edge from-layer="56" from-port="5" to-layer="61" to-port="0" />
-		<edge from-layer="57" from-port="0" to-layer="58" to-port="0" />
-		<edge from-layer="58" from-port="1" to-layer="60" to-port="0" />
-		<edge from-layer="59" from-port="0" to-layer="60" to-port="1" />
-		<edge from-layer="60" from-port="2" to-layer="61" to-port="1" />
-		<edge from-layer="61" from-port="2" to-layer="63" to-port="0" />
-		<edge from-layer="62" from-port="0" to-layer="63" to-port="1" />
-		<edge from-layer="63" from-port="2" to-layer="68" to-port="0" />
-		<edge from-layer="64" from-port="0" to-layer="68" to-port="1" />
-		<edge from-layer="65" from-port="0" to-layer="68" to-port="2" />
-		<edge from-layer="66" from-port="0" to-layer="68" to-port="3" />
-		<edge from-layer="67" from-port="0" to-layer="68" to-port="4" />
-		<edge from-layer="68" from-port="5" to-layer="70" to-port="0" />
-		<edge from-layer="69" from-port="0" to-layer="70" to-port="1" />
+		<edge from-layer="51" from-port="2" to-layer="52" to-port="0" />
+		<edge from-layer="52" from-port="1" to-layer="54" to-port="1" />
+		<edge from-layer="53" from-port="0" to-layer="54" to-port="2" />
+		<edge from-layer="54" from-port="3" to-layer="55" to-port="1" />
+		<edge from-layer="55" from-port="2" to-layer="57" to-port="0" />
+		<edge from-layer="56" from-port="0" to-layer="57" to-port="1" />
+		<edge from-layer="57" from-port="2" to-layer="59" to-port="0" />
+		<edge from-layer="58" from-port="0" to-layer="59" to-port="1" />
+		<edge from-layer="59" from-port="2" to-layer="61" to-port="0" />
+		<edge from-layer="60" from-port="0" to-layer="61" to-port="1" />
+		<edge from-layer="61" from-port="2" to-layer="70" to-port="0" />
+		<edge from-layer="61" from-port="2" to-layer="100" to-port="0" />
+		<edge from-layer="61" from-port="2" to-layer="141" to-port="1" />
+		<edge from-layer="61" from-port="2" to-layer="85" to-port="0" />
+		<edge from-layer="62" from-port="0" to-layer="63" to-port="0" />
+		<edge from-layer="63" from-port="1" to-layer="66" to-port="0" />
+		<edge from-layer="64" from-port="0" to-layer="65" to-port="0" />
+		<edge from-layer="65" from-port="1" to-layer="66" to-port="1" />
+		<edge from-layer="66" from-port="2" to-layer="68" to-port="0" />
+		<edge from-layer="67" from-port="0" to-layer="68" to-port="1" />
+		<edge from-layer="68" from-port="2" to-layer="69" to-port="0" />
+		<edge from-layer="69" from-port="1" to-layer="70" to-port="1" />
 		<edge from-layer="70" from-port="2" to-layer="72" to-port="0" />
 		<edge from-layer="71" from-port="0" to-layer="72" to-port="1" />
-		<edge from-layer="72" from-port="2" to-layer="120" to-port="0" />
-		<edge from-layer="73" from-port="0" to-layer="74" to-port="0" />
-		<edge from-layer="74" from-port="1" to-layer="76" to-port="0" />
+		<edge from-layer="72" from-port="2" to-layer="74" to-port="0" />
+		<edge from-layer="73" from-port="0" to-layer="74" to-port="1" />
+		<edge from-layer="74" from-port="2" to-layer="76" to-port="0" />
 		<edge from-layer="75" from-port="0" to-layer="76" to-port="1" />
-		<edge from-layer="76" from-port="2" to-layer="77" to-port="1" />
-		<edge from-layer="77" from-port="2" to-layer="79" to-port="0" />
-		<edge from-layer="78" from-port="0" to-layer="79" to-port="1" />
-		<edge from-layer="79" from-port="2" to-layer="84" to-port="0" />
-		<edge from-layer="80" from-port="0" to-layer="84" to-port="1" />
-		<edge from-layer="81" from-port="0" to-layer="84" to-port="2" />
-		<edge from-layer="82" from-port="0" to-layer="84" to-port="3" />
-		<edge from-layer="83" from-port="0" to-layer="84" to-port="4" />
-		<edge from-layer="84" from-port="5" to-layer="86" to-port="0" />
-		<edge from-layer="85" from-port="0" to-layer="86" to-port="1" />
-		<edge from-layer="86" from-port="2" to-layer="88" to-port="0" />
-		<edge from-layer="87" from-port="0" to-layer="88" to-port="1" />
-		<edge from-layer="88" from-port="2" to-layer="120" to-port="1" />
-		<edge from-layer="89" from-port="0" to-layer="90" to-port="0" />
-		<edge from-layer="90" from-port="1" to-layer="92" to-port="0" />
-		<edge from-layer="91" from-port="0" to-layer="92" to-port="1" />
-		<edge from-layer="92" from-port="2" to-layer="93" to-port="1" />
-		<edge from-layer="93" from-port="2" to-layer="95" to-port="0" />
-		<edge from-layer="94" from-port="0" to-layer="95" to-port="1" />
-		<edge from-layer="95" from-port="2" to-layer="97" to-port="0" />
-		<edge from-layer="96" from-port="0" to-layer="97" to-port="1" />
-		<edge from-layer="97" from-port="2" to-layer="99" to-port="0" />
-		<edge from-layer="98" from-port="0" to-layer="99" to-port="1" />
-		<edge from-layer="99" from-port="2" to-layer="120" to-port="2" />
-		<edge from-layer="100" from-port="0" to-layer="116" to-port="0" />
-		<edge from-layer="101" from-port="2" to-layer="103" to-port="0" />
-		<edge from-layer="102" from-port="0" to-layer="103" to-port="1" />
-		<edge from-layer="103" from-port="2" to-layer="112" to-port="0" />
-		<edge from-layer="104" from-port="0" to-layer="106" to-port="1" />
-		<edge from-layer="105" from-port="0" to-layer="106" to-port="2" />
-		<edge from-layer="106" from-port="3" to-layer="111" to-port="0" />
-		<edge from-layer="107" from-port="0" to-layer="111" to-port="1" />
-		<edge from-layer="108" from-port="0" to-layer="110" to-port="1" />
-		<edge from-layer="109" from-port="0" to-layer="110" to-port="2" />
-		<edge from-layer="110" from-port="3" to-layer="111" to-port="2" />
-		<edge from-layer="111" from-port="3" to-layer="112" to-port="1" />
-		<edge from-layer="112" from-port="2" to-layer="113" to-port="0" />
-		<edge from-layer="113" from-port="1" to-layer="115" to-port="0" />
-		<edge from-layer="114" from-port="0" to-layer="115" to-port="1" />
-		<edge from-layer="115" from-port="2" to-layer="116" to-port="1" />
-		<edge from-layer="116" from-port="2" to-layer="117" to-port="0" />
-		<edge from-layer="116" from-port="2" to-layer="119" to-port="2" />
-		<edge from-layer="117" from-port="1" to-layer="119" to-port="0" />
-		<edge from-layer="118" from-port="0" to-layer="119" to-port="1" />
-		<edge from-layer="119" from-port="3" to-layer="680" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="568" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="1352" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="1240" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="1128" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="1016" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="904" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="792" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="344" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="232" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="456" to-port="3" />
-		<edge from-layer="119" from-port="3" to-layer="120" to-port="3" />
-		<edge from-layer="120" from-port="4" to-layer="122" to-port="0" />
-		<edge from-layer="121" from-port="0" to-layer="122" to-port="1" />
-		<edge from-layer="122" from-port="2" to-layer="124" to-port="0" />
+		<edge from-layer="76" from-port="2" to-layer="125" to-port="0" />
+		<edge from-layer="77" from-port="0" to-layer="78" to-port="0" />
+		<edge from-layer="78" from-port="1" to-layer="81" to-port="0" />
+		<edge from-layer="79" from-port="0" to-layer="80" to-port="0" />
+		<edge from-layer="80" from-port="1" to-layer="81" to-port="1" />
+		<edge from-layer="81" from-port="2" to-layer="83" to-port="0" />
+		<edge from-layer="82" from-port="0" to-layer="83" to-port="1" />
+		<edge from-layer="83" from-port="2" to-layer="84" to-port="0" />
+		<edge from-layer="84" from-port="1" to-layer="85" to-port="1" />
+		<edge from-layer="85" from-port="2" to-layer="87" to-port="0" />
+		<edge from-layer="86" from-port="0" to-layer="87" to-port="1" />
+		<edge from-layer="87" from-port="2" to-layer="89" to-port="0" />
+		<edge from-layer="88" from-port="0" to-layer="89" to-port="1" />
+		<edge from-layer="89" from-port="2" to-layer="91" to-port="0" />
+		<edge from-layer="90" from-port="0" to-layer="91" to-port="1" />
+		<edge from-layer="91" from-port="2" to-layer="125" to-port="1" />
+		<edge from-layer="92" from-port="0" to-layer="93" to-port="0" />
+		<edge from-layer="93" from-port="1" to-layer="96" to-port="0" />
+		<edge from-layer="94" from-port="0" to-layer="95" to-port="0" />
+		<edge from-layer="95" from-port="1" to-layer="96" to-port="1" />
+		<edge from-layer="96" from-port="2" to-layer="98" to-port="0" />
+		<edge from-layer="97" from-port="0" to-layer="98" to-port="1" />
+		<edge from-layer="98" from-port="2" to-layer="99" to-port="0" />
+		<edge from-layer="99" from-port="1" to-layer="100" to-port="1" />
+		<edge from-layer="100" from-port="2" to-layer="102" to-port="0" />
+		<edge from-layer="101" from-port="0" to-layer="102" to-port="1" />
+		<edge from-layer="102" from-port="2" to-layer="104" to-port="0" />
+		<edge from-layer="103" from-port="0" to-layer="104" to-port="1" />
+		<edge from-layer="104" from-port="2" to-layer="106" to-port="0" />
+		<edge from-layer="105" from-port="0" to-layer="106" to-port="1" />
+		<edge from-layer="106" from-port="2" to-layer="125" to-port="2" />
+		<edge from-layer="107" from-port="0" to-layer="121" to-port="0" />
+		<edge from-layer="108" from-port="2" to-layer="110" to-port="0" />
+		<edge from-layer="109" from-port="0" to-layer="110" to-port="1" />
+		<edge from-layer="110" from-port="2" to-layer="119" to-port="0" />
+		<edge from-layer="111" from-port="0" to-layer="113" to-port="1" />
+		<edge from-layer="112" from-port="0" to-layer="113" to-port="2" />
+		<edge from-layer="113" from-port="3" to-layer="118" to-port="0" />
+		<edge from-layer="114" from-port="0" to-layer="118" to-port="1" />
+		<edge from-layer="115" from-port="0" to-layer="117" to-port="1" />
+		<edge from-layer="116" from-port="0" to-layer="117" to-port="2" />
+		<edge from-layer="117" from-port="3" to-layer="118" to-port="2" />
+		<edge from-layer="118" from-port="3" to-layer="119" to-port="1" />
+		<edge from-layer="119" from-port="2" to-layer="120" to-port="0" />
+		<edge from-layer="120" from-port="1" to-layer="121" to-port="1" />
+		<edge from-layer="121" from-port="2" to-layer="124" to-port="2" />
+		<edge from-layer="121" from-port="2" to-layer="122" to-port="0" />
+		<edge from-layer="122" from-port="1" to-layer="124" to-port="0" />
 		<edge from-layer="123" from-port="0" to-layer="124" to-port="1" />
-		<edge from-layer="124" from-port="2" to-layer="126" to-port="0" />
-		<edge from-layer="125" from-port="0" to-layer="126" to-port="1" />
-		<edge from-layer="126" from-port="2" to-layer="131" to-port="0" />
-		<edge from-layer="127" from-port="0" to-layer="131" to-port="1" />
-		<edge from-layer="128" from-port="0" to-layer="131" to-port="2" />
-		<edge from-layer="129" from-port="0" to-layer="131" to-port="3" />
-		<edge from-layer="130" from-port="0" to-layer="131" to-port="4" />
-		<edge from-layer="131" from-port="5" to-layer="136" to-port="0" />
+		<edge from-layer="124" from-port="3" to-layer="321" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="713" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="811" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="909" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="1007" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="1105" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="1203" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="615" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="517" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="419" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="223" to-port="3" />
+		<edge from-layer="124" from-port="3" to-layer="125" to-port="3" />
+		<edge from-layer="125" from-port="4" to-layer="127" to-port="0" />
+		<edge from-layer="126" from-port="0" to-layer="127" to-port="1" />
+		<edge from-layer="127" from-port="2" to-layer="129" to-port="0" />
+		<edge from-layer="128" from-port="0" to-layer="129" to-port="1" />
+		<edge from-layer="129" from-port="2" to-layer="138" to-port="0" />
+		<edge from-layer="130" from-port="0" to-layer="131" to-port="0" />
+		<edge from-layer="131" from-port="1" to-layer="134" to-port="0" />
 		<edge from-layer="132" from-port="0" to-layer="133" to-port="0" />
-		<edge from-layer="133" from-port="1" to-layer="135" to-port="0" />
-		<edge from-layer="134" from-port="0" to-layer="135" to-port="1" />
-		<edge from-layer="135" from-port="2" to-layer="136" to-port="1" />
-		<edge from-layer="136" from-port="2" to-layer="138" to-port="0" />
-		<edge from-layer="137" from-port="0" to-layer="138" to-port="1" />
-		<edge from-layer="138" from-port="2" to-layer="139" to-port="0" />
-		<edge from-layer="139" from-port="2" to-layer="141" to-port="0" />
-		<edge from-layer="140" from-port="0" to-layer="141" to-port="1" />
+		<edge from-layer="133" from-port="1" to-layer="134" to-port="1" />
+		<edge from-layer="134" from-port="2" to-layer="136" to-port="0" />
+		<edge from-layer="135" from-port="0" to-layer="136" to-port="1" />
+		<edge from-layer="136" from-port="2" to-layer="137" to-port="0" />
+		<edge from-layer="137" from-port="1" to-layer="138" to-port="1" />
+		<edge from-layer="138" from-port="2" to-layer="140" to-port="0" />
+		<edge from-layer="139" from-port="0" to-layer="140" to-port="1" />
+		<edge from-layer="140" from-port="2" to-layer="141" to-port="0" />
 		<edge from-layer="141" from-port="2" to-layer="143" to-port="0" />
 		<edge from-layer="142" from-port="0" to-layer="143" to-port="1" />
 		<edge from-layer="143" from-port="2" to-layer="145" to-port="0" />
 		<edge from-layer="144" from-port="0" to-layer="145" to-port="1" />
 		<edge from-layer="145" from-port="2" to-layer="147" to-port="0" />
-		<edge from-layer="145" from-port="2" to-layer="175" to-port="1" />
 		<edge from-layer="146" from-port="0" to-layer="147" to-port="1" />
-		<edge from-layer="147" from-port="2" to-layer="152" to-port="0" />
-		<edge from-layer="148" from-port="0" to-layer="152" to-port="1" />
-		<edge from-layer="149" from-port="0" to-layer="152" to-port="2" />
-		<edge from-layer="150" from-port="0" to-layer="152" to-port="3" />
-		<edge from-layer="151" from-port="0" to-layer="152" to-port="4" />
-		<edge from-layer="152" from-port="5" to-layer="157" to-port="0" />
-		<edge from-layer="153" from-port="0" to-layer="154" to-port="0" />
-		<edge from-layer="154" from-port="1" to-layer="156" to-port="0" />
-		<edge from-layer="155" from-port="0" to-layer="156" to-port="1" />
-		<edge from-layer="156" from-port="2" to-layer="157" to-port="1" />
-		<edge from-layer="157" from-port="2" to-layer="159" to-port="0" />
-		<edge from-layer="158" from-port="0" to-layer="159" to-port="1" />
-		<edge from-layer="159" from-port="2" to-layer="160" to-port="0" />
-		<edge from-layer="160" from-port="1" to-layer="162" to-port="0" />
-		<edge from-layer="161" from-port="0" to-layer="162" to-port="1" />
-		<edge from-layer="162" from-port="2" to-layer="167" to-port="0" />
-		<edge from-layer="163" from-port="0" to-layer="167" to-port="1" />
-		<edge from-layer="164" from-port="0" to-layer="167" to-port="2" />
-		<edge from-layer="165" from-port="0" to-layer="167" to-port="3" />
-		<edge from-layer="166" from-port="0" to-layer="167" to-port="4" />
-		<edge from-layer="167" from-port="5" to-layer="172" to-port="0" />
-		<edge from-layer="168" from-port="0" to-layer="169" to-port="0" />
-		<edge from-layer="169" from-port="1" to-layer="171" to-port="0" />
-		<edge from-layer="170" from-port="0" to-layer="171" to-port="1" />
-		<edge from-layer="171" from-port="2" to-layer="172" to-port="1" />
-		<edge from-layer="172" from-port="2" to-layer="174" to-port="0" />
-		<edge from-layer="173" from-port="0" to-layer="174" to-port="1" />
-		<edge from-layer="174" from-port="2" to-layer="175" to-port="0" />
+		<edge from-layer="147" from-port="2" to-layer="156" to-port="0" />
+		<edge from-layer="147" from-port="2" to-layer="171" to-port="1" />
+		<edge from-layer="148" from-port="0" to-layer="149" to-port="0" />
+		<edge from-layer="149" from-port="1" to-layer="152" to-port="0" />
+		<edge from-layer="150" from-port="0" to-layer="151" to-port="0" />
+		<edge from-layer="151" from-port="1" to-layer="152" to-port="1" />
+		<edge from-layer="152" from-port="2" to-layer="154" to-port="0" />
+		<edge from-layer="153" from-port="0" to-layer="154" to-port="1" />
+		<edge from-layer="154" from-port="2" to-layer="155" to-port="0" />
+		<edge from-layer="155" from-port="1" to-layer="156" to-port="1" />
+		<edge from-layer="156" from-port="2" to-layer="158" to-port="0" />
+		<edge from-layer="157" from-port="0" to-layer="158" to-port="1" />
+		<edge from-layer="158" from-port="2" to-layer="159" to-port="0" />
+		<edge from-layer="159" from-port="1" to-layer="168" to-port="0" />
+		<edge from-layer="160" from-port="0" to-layer="161" to-port="0" />
+		<edge from-layer="161" from-port="1" to-layer="164" to-port="0" />
+		<edge from-layer="162" from-port="0" to-layer="163" to-port="0" />
+		<edge from-layer="163" from-port="1" to-layer="164" to-port="1" />
+		<edge from-layer="164" from-port="2" to-layer="166" to-port="0" />
+		<edge from-layer="165" from-port="0" to-layer="166" to-port="1" />
+		<edge from-layer="166" from-port="2" to-layer="167" to-port="0" />
+		<edge from-layer="167" from-port="1" to-layer="168" to-port="1" />
+		<edge from-layer="168" from-port="2" to-layer="170" to-port="0" />
+		<edge from-layer="169" from-port="0" to-layer="170" to-port="1" />
+		<edge from-layer="170" from-port="2" to-layer="171" to-port="0" />
+		<edge from-layer="171" from-port="2" to-layer="173" to-port="0" />
+		<edge from-layer="172" from-port="0" to-layer="173" to-port="1" />
+		<edge from-layer="173" from-port="2" to-layer="175" to-port="0" />
+		<edge from-layer="174" from-port="0" to-layer="175" to-port="1" />
 		<edge from-layer="175" from-port="2" to-layer="177" to-port="0" />
 		<edge from-layer="176" from-port="0" to-layer="177" to-port="1" />
-		<edge from-layer="177" from-port="2" to-layer="179" to-port="0" />
-		<edge from-layer="178" from-port="0" to-layer="179" to-port="1" />
-		<edge from-layer="179" from-port="2" to-layer="181" to-port="0" />
-		<edge from-layer="180" from-port="0" to-layer="181" to-port="1" />
-		<edge from-layer="181" from-port="2" to-layer="251" to-port="1" />
-		<edge from-layer="181" from-port="2" to-layer="183" to-port="0" />
-		<edge from-layer="182" from-port="0" to-layer="183" to-port="1" />
-		<edge from-layer="183" from-port="2" to-layer="188" to-port="0" />
-		<edge from-layer="184" from-port="0" to-layer="188" to-port="1" />
-		<edge from-layer="185" from-port="0" to-layer="188" to-port="2" />
-		<edge from-layer="186" from-port="0" to-layer="188" to-port="3" />
-		<edge from-layer="187" from-port="0" to-layer="188" to-port="4" />
-		<edge from-layer="188" from-port="5" to-layer="225" to-port="0" />
-		<edge from-layer="188" from-port="5" to-layer="193" to-port="0" />
-		<edge from-layer="188" from-port="5" to-layer="209" to-port="0" />
-		<edge from-layer="189" from-port="0" to-layer="190" to-port="0" />
-		<edge from-layer="190" from-port="1" to-layer="192" to-port="0" />
+		<edge from-layer="177" from-port="2" to-layer="239" to-port="1" />
+		<edge from-layer="177" from-port="2" to-layer="186" to-port="0" />
+		<edge from-layer="177" from-port="2" to-layer="216" to-port="0" />
+		<edge from-layer="177" from-port="2" to-layer="201" to-port="0" />
+		<edge from-layer="178" from-port="0" to-layer="179" to-port="0" />
+		<edge from-layer="179" from-port="1" to-layer="182" to-port="0" />
+		<edge from-layer="180" from-port="0" to-layer="181" to-port="0" />
+		<edge from-layer="181" from-port="1" to-layer="182" to-port="1" />
+		<edge from-layer="182" from-port="2" to-layer="184" to-port="0" />
+		<edge from-layer="183" from-port="0" to-layer="184" to-port="1" />
+		<edge from-layer="184" from-port="2" to-layer="185" to-port="0" />
+		<edge from-layer="185" from-port="1" to-layer="186" to-port="1" />
+		<edge from-layer="186" from-port="2" to-layer="188" to-port="0" />
+		<edge from-layer="187" from-port="0" to-layer="188" to-port="1" />
+		<edge from-layer="188" from-port="2" to-layer="190" to-port="0" />
+		<edge from-layer="189" from-port="0" to-layer="190" to-port="1" />
+		<edge from-layer="190" from-port="2" to-layer="192" to-port="0" />
 		<edge from-layer="191" from-port="0" to-layer="192" to-port="1" />
-		<edge from-layer="192" from-port="2" to-layer="193" to-port="1" />
-		<edge from-layer="193" from-port="2" to-layer="195" to-port="0" />
-		<edge from-layer="194" from-port="0" to-layer="195" to-port="1" />
-		<edge from-layer="195" from-port="2" to-layer="200" to-port="0" />
-		<edge from-layer="196" from-port="0" to-layer="200" to-port="1" />
-		<edge from-layer="197" from-port="0" to-layer="200" to-port="2" />
-		<edge from-layer="198" from-port="0" to-layer="200" to-port="3" />
-		<edge from-layer="199" from-port="0" to-layer="200" to-port="4" />
-		<edge from-layer="200" from-port="5" to-layer="202" to-port="0" />
-		<edge from-layer="201" from-port="0" to-layer="202" to-port="1" />
-		<edge from-layer="202" from-port="2" to-layer="204" to-port="0" />
-		<edge from-layer="203" from-port="0" to-layer="204" to-port="1" />
-		<edge from-layer="204" from-port="2" to-layer="232" to-port="0" />
-		<edge from-layer="205" from-port="0" to-layer="206" to-port="0" />
-		<edge from-layer="206" from-port="1" to-layer="208" to-port="0" />
-		<edge from-layer="207" from-port="0" to-layer="208" to-port="1" />
-		<edge from-layer="208" from-port="2" to-layer="209" to-port="1" />
-		<edge from-layer="209" from-port="2" to-layer="211" to-port="0" />
-		<edge from-layer="210" from-port="0" to-layer="211" to-port="1" />
-		<edge from-layer="211" from-port="2" to-layer="216" to-port="0" />
-		<edge from-layer="212" from-port="0" to-layer="216" to-port="1" />
-		<edge from-layer="213" from-port="0" to-layer="216" to-port="2" />
-		<edge from-layer="214" from-port="0" to-layer="216" to-port="3" />
-		<edge from-layer="215" from-port="0" to-layer="216" to-port="4" />
-		<edge from-layer="216" from-port="5" to-layer="218" to-port="0" />
+		<edge from-layer="192" from-port="2" to-layer="223" to-port="0" />
+		<edge from-layer="193" from-port="0" to-layer="194" to-port="0" />
+		<edge from-layer="194" from-port="1" to-layer="197" to-port="0" />
+		<edge from-layer="195" from-port="0" to-layer="196" to-port="0" />
+		<edge from-layer="196" from-port="1" to-layer="197" to-port="1" />
+		<edge from-layer="197" from-port="2" to-layer="199" to-port="0" />
+		<edge from-layer="198" from-port="0" to-layer="199" to-port="1" />
+		<edge from-layer="199" from-port="2" to-layer="200" to-port="0" />
+		<edge from-layer="200" from-port="1" to-layer="201" to-port="1" />
+		<edge from-layer="201" from-port="2" to-layer="203" to-port="0" />
+		<edge from-layer="202" from-port="0" to-layer="203" to-port="1" />
+		<edge from-layer="203" from-port="2" to-layer="205" to-port="0" />
+		<edge from-layer="204" from-port="0" to-layer="205" to-port="1" />
+		<edge from-layer="205" from-port="2" to-layer="207" to-port="0" />
+		<edge from-layer="206" from-port="0" to-layer="207" to-port="1" />
+		<edge from-layer="207" from-port="2" to-layer="223" to-port="1" />
+		<edge from-layer="208" from-port="0" to-layer="209" to-port="0" />
+		<edge from-layer="209" from-port="1" to-layer="212" to-port="0" />
+		<edge from-layer="210" from-port="0" to-layer="211" to-port="0" />
+		<edge from-layer="211" from-port="1" to-layer="212" to-port="1" />
+		<edge from-layer="212" from-port="2" to-layer="214" to-port="0" />
+		<edge from-layer="213" from-port="0" to-layer="214" to-port="1" />
+		<edge from-layer="214" from-port="2" to-layer="215" to-port="0" />
+		<edge from-layer="215" from-port="1" to-layer="216" to-port="1" />
+		<edge from-layer="216" from-port="2" to-layer="218" to-port="0" />
 		<edge from-layer="217" from-port="0" to-layer="218" to-port="1" />
 		<edge from-layer="218" from-port="2" to-layer="220" to-port="0" />
 		<edge from-layer="219" from-port="0" to-layer="220" to-port="1" />
-		<edge from-layer="220" from-port="2" to-layer="232" to-port="1" />
-		<edge from-layer="221" from-port="0" to-layer="222" to-port="0" />
-		<edge from-layer="222" from-port="1" to-layer="224" to-port="0" />
-		<edge from-layer="223" from-port="0" to-layer="224" to-port="1" />
-		<edge from-layer="224" from-port="2" to-layer="225" to-port="1" />
+		<edge from-layer="220" from-port="2" to-layer="222" to-port="0" />
+		<edge from-layer="221" from-port="0" to-layer="222" to-port="1" />
+		<edge from-layer="222" from-port="2" to-layer="223" to-port="2" />
+		<edge from-layer="223" from-port="4" to-layer="225" to-port="0" />
+		<edge from-layer="224" from-port="0" to-layer="225" to-port="1" />
 		<edge from-layer="225" from-port="2" to-layer="227" to-port="0" />
 		<edge from-layer="226" from-port="0" to-layer="227" to-port="1" />
-		<edge from-layer="227" from-port="2" to-layer="229" to-port="0" />
-		<edge from-layer="228" from-port="0" to-layer="229" to-port="1" />
-		<edge from-layer="229" from-port="2" to-layer="231" to-port="0" />
-		<edge from-layer="230" from-port="0" to-layer="231" to-port="1" />
-		<edge from-layer="231" from-port="2" to-layer="232" to-port="2" />
-		<edge from-layer="232" from-port="4" to-layer="234" to-port="0" />
+		<edge from-layer="227" from-port="2" to-layer="236" to-port="0" />
+		<edge from-layer="228" from-port="0" to-layer="229" to-port="0" />
+		<edge from-layer="229" from-port="1" to-layer="232" to-port="0" />
+		<edge from-layer="230" from-port="0" to-layer="231" to-port="0" />
+		<edge from-layer="231" from-port="1" to-layer="232" to-port="1" />
+		<edge from-layer="232" from-port="2" to-layer="234" to-port="0" />
 		<edge from-layer="233" from-port="0" to-layer="234" to-port="1" />
-		<edge from-layer="234" from-port="2" to-layer="236" to-port="0" />
-		<edge from-layer="235" from-port="0" to-layer="236" to-port="1" />
+		<edge from-layer="234" from-port="2" to-layer="235" to-port="0" />
+		<edge from-layer="235" from-port="1" to-layer="236" to-port="1" />
 		<edge from-layer="236" from-port="2" to-layer="238" to-port="0" />
 		<edge from-layer="237" from-port="0" to-layer="238" to-port="1" />
-		<edge from-layer="238" from-port="2" to-layer="243" to-port="0" />
-		<edge from-layer="239" from-port="0" to-layer="243" to-port="1" />
-		<edge from-layer="240" from-port="0" to-layer="243" to-port="2" />
-		<edge from-layer="241" from-port="0" to-layer="243" to-port="3" />
-		<edge from-layer="242" from-port="0" to-layer="243" to-port="4" />
-		<edge from-layer="243" from-port="5" to-layer="248" to-port="0" />
-		<edge from-layer="244" from-port="0" to-layer="245" to-port="0" />
-		<edge from-layer="245" from-port="1" to-layer="247" to-port="0" />
-		<edge from-layer="246" from-port="0" to-layer="247" to-port="1" />
-		<edge from-layer="247" from-port="2" to-layer="248" to-port="1" />
-		<edge from-layer="248" from-port="2" to-layer="250" to-port="0" />
-		<edge from-layer="249" from-port="0" to-layer="250" to-port="1" />
-		<edge from-layer="250" from-port="2" to-layer="251" to-port="0" />
-		<edge from-layer="251" from-port="2" to-layer="253" to-port="0" />
-		<edge from-layer="252" from-port="0" to-layer="253" to-port="1" />
-		<edge from-layer="253" from-port="2" to-layer="255" to-port="0" />
-		<edge from-layer="254" from-port="0" to-layer="255" to-port="1" />
-		<edge from-layer="255" from-port="2" to-layer="257" to-port="0" />
-		<edge from-layer="256" from-port="0" to-layer="257" to-port="1" />
-		<edge from-layer="257" from-port="2" to-layer="259" to-port="0" />
-		<edge from-layer="257" from-port="2" to-layer="287" to-port="1" />
-		<edge from-layer="258" from-port="0" to-layer="259" to-port="1" />
-		<edge from-layer="259" from-port="2" to-layer="264" to-port="0" />
-		<edge from-layer="260" from-port="0" to-layer="264" to-port="1" />
-		<edge from-layer="261" from-port="0" to-layer="264" to-port="2" />
-		<edge from-layer="262" from-port="0" to-layer="264" to-port="3" />
-		<edge from-layer="263" from-port="0" to-layer="264" to-port="4" />
-		<edge from-layer="264" from-port="5" to-layer="269" to-port="0" />
-		<edge from-layer="265" from-port="0" to-layer="266" to-port="0" />
-		<edge from-layer="266" from-port="1" to-layer="268" to-port="0" />
+		<edge from-layer="238" from-port="2" to-layer="239" to-port="0" />
+		<edge from-layer="239" from-port="2" to-layer="241" to-port="0" />
+		<edge from-layer="240" from-port="0" to-layer="241" to-port="1" />
+		<edge from-layer="241" from-port="2" to-layer="243" to-port="0" />
+		<edge from-layer="242" from-port="0" to-layer="243" to-port="1" />
+		<edge from-layer="243" from-port="2" to-layer="245" to-port="0" />
+		<edge from-layer="244" from-port="0" to-layer="245" to-port="1" />
+		<edge from-layer="245" from-port="2" to-layer="269" to-port="1" />
+		<edge from-layer="245" from-port="2" to-layer="254" to-port="0" />
+		<edge from-layer="246" from-port="0" to-layer="247" to-port="0" />
+		<edge from-layer="247" from-port="1" to-layer="250" to-port="0" />
+		<edge from-layer="248" from-port="0" to-layer="249" to-port="0" />
+		<edge from-layer="249" from-port="1" to-layer="250" to-port="1" />
+		<edge from-layer="250" from-port="2" to-layer="252" to-port="0" />
+		<edge from-layer="251" from-port="0" to-layer="252" to-port="1" />
+		<edge from-layer="252" from-port="2" to-layer="253" to-port="0" />
+		<edge from-layer="253" from-port="1" to-layer="254" to-port="1" />
+		<edge from-layer="254" from-port="2" to-layer="256" to-port="0" />
+		<edge from-layer="255" from-port="0" to-layer="256" to-port="1" />
+		<edge from-layer="256" from-port="2" to-layer="257" to-port="0" />
+		<edge from-layer="257" from-port="1" to-layer="266" to-port="0" />
+		<edge from-layer="258" from-port="0" to-layer="259" to-port="0" />
+		<edge from-layer="259" from-port="1" to-layer="262" to-port="0" />
+		<edge from-layer="260" from-port="0" to-layer="261" to-port="0" />
+		<edge from-layer="261" from-port="1" to-layer="262" to-port="1" />
+		<edge from-layer="262" from-port="2" to-layer="264" to-port="0" />
+		<edge from-layer="263" from-port="0" to-layer="264" to-port="1" />
+		<edge from-layer="264" from-port="2" to-layer="265" to-port="0" />
+		<edge from-layer="265" from-port="1" to-layer="266" to-port="1" />
+		<edge from-layer="266" from-port="2" to-layer="268" to-port="0" />
 		<edge from-layer="267" from-port="0" to-layer="268" to-port="1" />
-		<edge from-layer="268" from-port="2" to-layer="269" to-port="1" />
+		<edge from-layer="268" from-port="2" to-layer="269" to-port="0" />
 		<edge from-layer="269" from-port="2" to-layer="271" to-port="0" />
 		<edge from-layer="270" from-port="0" to-layer="271" to-port="1" />
-		<edge from-layer="271" from-port="2" to-layer="272" to-port="0" />
-		<edge from-layer="272" from-port="1" to-layer="274" to-port="0" />
-		<edge from-layer="273" from-port="0" to-layer="274" to-port="1" />
-		<edge from-layer="274" from-port="2" to-layer="279" to-port="0" />
-		<edge from-layer="275" from-port="0" to-layer="279" to-port="1" />
-		<edge from-layer="276" from-port="0" to-layer="279" to-port="2" />
-		<edge from-layer="277" from-port="0" to-layer="279" to-port="3" />
-		<edge from-layer="278" from-port="0" to-layer="279" to-port="4" />
-		<edge from-layer="279" from-port="5" to-layer="284" to-port="0" />
-		<edge from-layer="280" from-port="0" to-layer="281" to-port="0" />
-		<edge from-layer="281" from-port="1" to-layer="283" to-port="0" />
-		<edge from-layer="282" from-port="0" to-layer="283" to-port="1" />
-		<edge from-layer="283" from-port="2" to-layer="284" to-port="1" />
+		<edge from-layer="271" from-port="2" to-layer="273" to-port="0" />
+		<edge from-layer="272" from-port="0" to-layer="273" to-port="1" />
+		<edge from-layer="273" from-port="2" to-layer="275" to-port="0" />
+		<edge from-layer="274" from-port="0" to-layer="275" to-port="1" />
+		<edge from-layer="275" from-port="2" to-layer="337" to-port="1" />
+		<edge from-layer="275" from-port="2" to-layer="314" to-port="0" />
+		<edge from-layer="275" from-port="2" to-layer="299" to-port="0" />
+		<edge from-layer="275" from-port="2" to-layer="284" to-port="0" />
+		<edge from-layer="276" from-port="0" to-layer="277" to-port="0" />
+		<edge from-layer="277" from-port="1" to-layer="280" to-port="0" />
+		<edge from-layer="278" from-port="0" to-layer="279" to-port="0" />
+		<edge from-layer="279" from-port="1" to-layer="280" to-port="1" />
+		<edge from-layer="280" from-port="2" to-layer="282" to-port="0" />
+		<edge from-layer="281" from-port="0" to-layer="282" to-port="1" />
+		<edge from-layer="282" from-port="2" to-layer="283" to-port="0" />
+		<edge from-layer="283" from-port="1" to-layer="284" to-port="1" />
 		<edge from-layer="284" from-port="2" to-layer="286" to-port="0" />
 		<edge from-layer="285" from-port="0" to-layer="286" to-port="1" />
-		<edge from-layer="286" from-port="2" to-layer="287" to-port="0" />
-		<edge from-layer="287" from-port="2" to-layer="289" to-port="0" />
-		<edge from-layer="288" from-port="0" to-layer="289" to-port="1" />
-		<edge from-layer="289" from-port="2" to-layer="291" to-port="0" />
-		<edge from-layer="290" from-port="0" to-layer="291" to-port="1" />
-		<edge from-layer="291" from-port="2" to-layer="293" to-port="0" />
-		<edge from-layer="292" from-port="0" to-layer="293" to-port="1" />
-		<edge from-layer="293" from-port="2" to-layer="363" to-port="1" />
-		<edge from-layer="293" from-port="2" to-layer="295" to-port="0" />
-		<edge from-layer="294" from-port="0" to-layer="295" to-port="1" />
-		<edge from-layer="295" from-port="2" to-layer="300" to-port="0" />
-		<edge from-layer="296" from-port="0" to-layer="300" to-port="1" />
-		<edge from-layer="297" from-port="0" to-layer="300" to-port="2" />
-		<edge from-layer="298" from-port="0" to-layer="300" to-port="3" />
-		<edge from-layer="299" from-port="0" to-layer="300" to-port="4" />
-		<edge from-layer="300" from-port="5" to-layer="337" to-port="0" />
-		<edge from-layer="300" from-port="5" to-layer="305" to-port="0" />
-		<edge from-layer="300" from-port="5" to-layer="321" to-port="0" />
-		<edge from-layer="301" from-port="0" to-layer="302" to-port="0" />
-		<edge from-layer="302" from-port="1" to-layer="304" to-port="0" />
-		<edge from-layer="303" from-port="0" to-layer="304" to-port="1" />
-		<edge from-layer="304" from-port="2" to-layer="305" to-port="1" />
-		<edge from-layer="305" from-port="2" to-layer="307" to-port="0" />
-		<edge from-layer="306" from-port="0" to-layer="307" to-port="1" />
-		<edge from-layer="307" from-port="2" to-layer="312" to-port="0" />
-		<edge from-layer="308" from-port="0" to-layer="312" to-port="1" />
-		<edge from-layer="309" from-port="0" to-layer="312" to-port="2" />
-		<edge from-layer="310" from-port="0" to-layer="312" to-port="3" />
-		<edge from-layer="311" from-port="0" to-layer="312" to-port="4" />
-		<edge from-layer="312" from-port="5" to-layer="314" to-port="0" />
-		<edge from-layer="313" from-port="0" to-layer="314" to-port="1" />
+		<edge from-layer="286" from-port="2" to-layer="288" to-port="0" />
+		<edge from-layer="287" from-port="0" to-layer="288" to-port="1" />
+		<edge from-layer="288" from-port="2" to-layer="290" to-port="0" />
+		<edge from-layer="289" from-port="0" to-layer="290" to-port="1" />
+		<edge from-layer="290" from-port="2" to-layer="321" to-port="0" />
+		<edge from-layer="291" from-port="0" to-layer="292" to-port="0" />
+		<edge from-layer="292" from-port="1" to-layer="295" to-port="0" />
+		<edge from-layer="293" from-port="0" to-layer="294" to-port="0" />
+		<edge from-layer="294" from-port="1" to-layer="295" to-port="1" />
+		<edge from-layer="295" from-port="2" to-layer="297" to-port="0" />
+		<edge from-layer="296" from-port="0" to-layer="297" to-port="1" />
+		<edge from-layer="297" from-port="2" to-layer="298" to-port="0" />
+		<edge from-layer="298" from-port="1" to-layer="299" to-port="1" />
+		<edge from-layer="299" from-port="2" to-layer="301" to-port="0" />
+		<edge from-layer="300" from-port="0" to-layer="301" to-port="1" />
+		<edge from-layer="301" from-port="2" to-layer="303" to-port="0" />
+		<edge from-layer="302" from-port="0" to-layer="303" to-port="1" />
+		<edge from-layer="303" from-port="2" to-layer="305" to-port="0" />
+		<edge from-layer="304" from-port="0" to-layer="305" to-port="1" />
+		<edge from-layer="305" from-port="2" to-layer="321" to-port="1" />
+		<edge from-layer="306" from-port="0" to-layer="307" to-port="0" />
+		<edge from-layer="307" from-port="1" to-layer="310" to-port="0" />
+		<edge from-layer="308" from-port="0" to-layer="309" to-port="0" />
+		<edge from-layer="309" from-port="1" to-layer="310" to-port="1" />
+		<edge from-layer="310" from-port="2" to-layer="312" to-port="0" />
+		<edge from-layer="311" from-port="0" to-layer="312" to-port="1" />
+		<edge from-layer="312" from-port="2" to-layer="313" to-port="0" />
+		<edge from-layer="313" from-port="1" to-layer="314" to-port="1" />
 		<edge from-layer="314" from-port="2" to-layer="316" to-port="0" />
 		<edge from-layer="315" from-port="0" to-layer="316" to-port="1" />
-		<edge from-layer="316" from-port="2" to-layer="344" to-port="0" />
-		<edge from-layer="317" from-port="0" to-layer="318" to-port="0" />
-		<edge from-layer="318" from-port="1" to-layer="320" to-port="0" />
+		<edge from-layer="316" from-port="2" to-layer="318" to-port="0" />
+		<edge from-layer="317" from-port="0" to-layer="318" to-port="1" />
+		<edge from-layer="318" from-port="2" to-layer="320" to-port="0" />
 		<edge from-layer="319" from-port="0" to-layer="320" to-port="1" />
-		<edge from-layer="320" from-port="2" to-layer="321" to-port="1" />
-		<edge from-layer="321" from-port="2" to-layer="323" to-port="0" />
+		<edge from-layer="320" from-port="2" to-layer="321" to-port="2" />
+		<edge from-layer="321" from-port="4" to-layer="323" to-port="0" />
 		<edge from-layer="322" from-port="0" to-layer="323" to-port="1" />
-		<edge from-layer="323" from-port="2" to-layer="328" to-port="0" />
-		<edge from-layer="324" from-port="0" to-layer="328" to-port="1" />
-		<edge from-layer="325" from-port="0" to-layer="328" to-port="2" />
-		<edge from-layer="326" from-port="0" to-layer="328" to-port="3" />
-		<edge from-layer="327" from-port="0" to-layer="328" to-port="4" />
-		<edge from-layer="328" from-port="5" to-layer="330" to-port="0" />
-		<edge from-layer="329" from-port="0" to-layer="330" to-port="1" />
+		<edge from-layer="323" from-port="2" to-layer="325" to-port="0" />
+		<edge from-layer="324" from-port="0" to-layer="325" to-port="1" />
+		<edge from-layer="325" from-port="2" to-layer="334" to-port="0" />
+		<edge from-layer="326" from-port="0" to-layer="327" to-port="0" />
+		<edge from-layer="327" from-port="1" to-layer="330" to-port="0" />
+		<edge from-layer="328" from-port="0" to-layer="329" to-port="0" />
+		<edge from-layer="329" from-port="1" to-layer="330" to-port="1" />
 		<edge from-layer="330" from-port="2" to-layer="332" to-port="0" />
 		<edge from-layer="331" from-port="0" to-layer="332" to-port="1" />
-		<edge from-layer="332" from-port="2" to-layer="344" to-port="1" />
-		<edge from-layer="333" from-port="0" to-layer="334" to-port="0" />
-		<edge from-layer="334" from-port="1" to-layer="336" to-port="0" />
+		<edge from-layer="332" from-port="2" to-layer="333" to-port="0" />
+		<edge from-layer="333" from-port="1" to-layer="334" to-port="1" />
+		<edge from-layer="334" from-port="2" to-layer="336" to-port="0" />
 		<edge from-layer="335" from-port="0" to-layer="336" to-port="1" />
-		<edge from-layer="336" from-port="2" to-layer="337" to-port="1" />
+		<edge from-layer="336" from-port="2" to-layer="337" to-port="0" />
 		<edge from-layer="337" from-port="2" to-layer="339" to-port="0" />
 		<edge from-layer="338" from-port="0" to-layer="339" to-port="1" />
 		<edge from-layer="339" from-port="2" to-layer="341" to-port="0" />
 		<edge from-layer="340" from-port="0" to-layer="341" to-port="1" />
 		<edge from-layer="341" from-port="2" to-layer="343" to-port="0" />
 		<edge from-layer="342" from-port="0" to-layer="343" to-port="1" />
-		<edge from-layer="343" from-port="2" to-layer="344" to-port="2" />
-		<edge from-layer="344" from-port="4" to-layer="346" to-port="0" />
-		<edge from-layer="345" from-port="0" to-layer="346" to-port="1" />
-		<edge from-layer="346" from-port="2" to-layer="348" to-port="0" />
-		<edge from-layer="347" from-port="0" to-layer="348" to-port="1" />
+		<edge from-layer="343" from-port="2" to-layer="367" to-port="1" />
+		<edge from-layer="343" from-port="2" to-layer="352" to-port="0" />
+		<edge from-layer="344" from-port="0" to-layer="345" to-port="0" />
+		<edge from-layer="345" from-port="1" to-layer="348" to-port="0" />
+		<edge from-layer="346" from-port="0" to-layer="347" to-port="0" />
+		<edge from-layer="347" from-port="1" to-layer="348" to-port="1" />
 		<edge from-layer="348" from-port="2" to-layer="350" to-port="0" />
 		<edge from-layer="349" from-port="0" to-layer="350" to-port="1" />
-		<edge from-layer="350" from-port="2" to-layer="355" to-port="0" />
-		<edge from-layer="351" from-port="0" to-layer="355" to-port="1" />
-		<edge from-layer="352" from-port="0" to-layer="355" to-port="2" />
-		<edge from-layer="353" from-port="0" to-layer="355" to-port="3" />
-		<edge from-layer="354" from-port="0" to-layer="355" to-port="4" />
-		<edge from-layer="355" from-port="5" to-layer="360" to-port="0" />
+		<edge from-layer="350" from-port="2" to-layer="351" to-port="0" />
+		<edge from-layer="351" from-port="1" to-layer="352" to-port="1" />
+		<edge from-layer="352" from-port="2" to-layer="354" to-port="0" />
+		<edge from-layer="353" from-port="0" to-layer="354" to-port="1" />
+		<edge from-layer="354" from-port="2" to-layer="355" to-port="0" />
+		<edge from-layer="355" from-port="1" to-layer="364" to-port="0" />
 		<edge from-layer="356" from-port="0" to-layer="357" to-port="0" />
-		<edge from-layer="357" from-port="1" to-layer="359" to-port="0" />
-		<edge from-layer="358" from-port="0" to-layer="359" to-port="1" />
-		<edge from-layer="359" from-port="2" to-layer="360" to-port="1" />
+		<edge from-layer="357" from-port="1" to-layer="360" to-port="0" />
+		<edge from-layer="358" from-port="0" to-layer="359" to-port="0" />
+		<edge from-layer="359" from-port="1" to-layer="360" to-port="1" />
 		<edge from-layer="360" from-port="2" to-layer="362" to-port="0" />
 		<edge from-layer="361" from-port="0" to-layer="362" to-port="1" />
 		<edge from-layer="362" from-port="2" to-layer="363" to-port="0" />
-		<edge from-layer="363" from-port="2" to-layer="365" to-port="0" />
-		<edge from-layer="364" from-port="0" to-layer="365" to-port="1" />
-		<edge from-layer="365" from-port="2" to-layer="367" to-port="0" />
-		<edge from-layer="366" from-port="0" to-layer="367" to-port="1" />
+		<edge from-layer="363" from-port="1" to-layer="364" to-port="1" />
+		<edge from-layer="364" from-port="2" to-layer="366" to-port="0" />
+		<edge from-layer="365" from-port="0" to-layer="366" to-port="1" />
+		<edge from-layer="366" from-port="2" to-layer="367" to-port="0" />
 		<edge from-layer="367" from-port="2" to-layer="369" to-port="0" />
 		<edge from-layer="368" from-port="0" to-layer="369" to-port="1" />
 		<edge from-layer="369" from-port="2" to-layer="371" to-port="0" />
-		<edge from-layer="369" from-port="2" to-layer="399" to-port="1" />
 		<edge from-layer="370" from-port="0" to-layer="371" to-port="1" />
-		<edge from-layer="371" from-port="2" to-layer="376" to-port="0" />
-		<edge from-layer="372" from-port="0" to-layer="376" to-port="1" />
-		<edge from-layer="373" from-port="0" to-layer="376" to-port="2" />
-		<edge from-layer="374" from-port="0" to-layer="376" to-port="3" />
-		<edge from-layer="375" from-port="0" to-layer="376" to-port="4" />
-		<edge from-layer="376" from-port="5" to-layer="381" to-port="0" />
-		<edge from-layer="377" from-port="0" to-layer="378" to-port="0" />
-		<edge from-layer="378" from-port="1" to-layer="380" to-port="0" />
+		<edge from-layer="371" from-port="2" to-layer="373" to-port="0" />
+		<edge from-layer="372" from-port="0" to-layer="373" to-port="1" />
+		<edge from-layer="373" from-port="2" to-layer="435" to-port="1" />
+		<edge from-layer="373" from-port="2" to-layer="412" to-port="0" />
+		<edge from-layer="373" from-port="2" to-layer="397" to-port="0" />
+		<edge from-layer="373" from-port="2" to-layer="382" to-port="0" />
+		<edge from-layer="374" from-port="0" to-layer="375" to-port="0" />
+		<edge from-layer="375" from-port="1" to-layer="378" to-port="0" />
+		<edge from-layer="376" from-port="0" to-layer="377" to-port="0" />
+		<edge from-layer="377" from-port="1" to-layer="378" to-port="1" />
+		<edge from-layer="378" from-port="2" to-layer="380" to-port="0" />
 		<edge from-layer="379" from-port="0" to-layer="380" to-port="1" />
-		<edge from-layer="380" from-port="2" to-layer="381" to-port="1" />
-		<edge from-layer="381" from-port="2" to-layer="383" to-port="0" />
-		<edge from-layer="382" from-port="0" to-layer="383" to-port="1" />
-		<edge from-layer="383" from-port="2" to-layer="384" to-port="0" />
-		<edge from-layer="384" from-port="1" to-layer="386" to-port="0" />
+		<edge from-layer="380" from-port="2" to-layer="381" to-port="0" />
+		<edge from-layer="381" from-port="1" to-layer="382" to-port="1" />
+		<edge from-layer="382" from-port="2" to-layer="384" to-port="0" />
+		<edge from-layer="383" from-port="0" to-layer="384" to-port="1" />
+		<edge from-layer="384" from-port="2" to-layer="386" to-port="0" />
 		<edge from-layer="385" from-port="0" to-layer="386" to-port="1" />
-		<edge from-layer="386" from-port="2" to-layer="391" to-port="0" />
-		<edge from-layer="387" from-port="0" to-layer="391" to-port="1" />
-		<edge from-layer="388" from-port="0" to-layer="391" to-port="2" />
-		<edge from-layer="389" from-port="0" to-layer="391" to-port="3" />
-		<edge from-layer="390" from-port="0" to-layer="391" to-port="4" />
-		<edge from-layer="391" from-port="5" to-layer="396" to-port="0" />
-		<edge from-layer="392" from-port="0" to-layer="393" to-port="0" />
-		<edge from-layer="393" from-port="1" to-layer="395" to-port="0" />
+		<edge from-layer="386" from-port="2" to-layer="388" to-port="0" />
+		<edge from-layer="387" from-port="0" to-layer="388" to-port="1" />
+		<edge from-layer="388" from-port="2" to-layer="419" to-port="0" />
+		<edge from-layer="389" from-port="0" to-layer="390" to-port="0" />
+		<edge from-layer="390" from-port="1" to-layer="393" to-port="0" />
+		<edge from-layer="391" from-port="0" to-layer="392" to-port="0" />
+		<edge from-layer="392" from-port="1" to-layer="393" to-port="1" />
+		<edge from-layer="393" from-port="2" to-layer="395" to-port="0" />
 		<edge from-layer="394" from-port="0" to-layer="395" to-port="1" />
-		<edge from-layer="395" from-port="2" to-layer="396" to-port="1" />
-		<edge from-layer="396" from-port="2" to-layer="398" to-port="0" />
-		<edge from-layer="397" from-port="0" to-layer="398" to-port="1" />
-		<edge from-layer="398" from-port="2" to-layer="399" to-port="0" />
+		<edge from-layer="395" from-port="2" to-layer="396" to-port="0" />
+		<edge from-layer="396" from-port="1" to-layer="397" to-port="1" />
+		<edge from-layer="397" from-port="2" to-layer="399" to-port="0" />
+		<edge from-layer="398" from-port="0" to-layer="399" to-port="1" />
 		<edge from-layer="399" from-port="2" to-layer="401" to-port="0" />
 		<edge from-layer="400" from-port="0" to-layer="401" to-port="1" />
 		<edge from-layer="401" from-port="2" to-layer="403" to-port="0" />
 		<edge from-layer="402" from-port="0" to-layer="403" to-port="1" />
-		<edge from-layer="403" from-port="2" to-layer="405" to-port="0" />
-		<edge from-layer="404" from-port="0" to-layer="405" to-port="1" />
-		<edge from-layer="405" from-port="2" to-layer="475" to-port="1" />
-		<edge from-layer="405" from-port="2" to-layer="407" to-port="0" />
-		<edge from-layer="406" from-port="0" to-layer="407" to-port="1" />
-		<edge from-layer="407" from-port="2" to-layer="412" to-port="0" />
-		<edge from-layer="408" from-port="0" to-layer="412" to-port="1" />
-		<edge from-layer="409" from-port="0" to-layer="412" to-port="2" />
-		<edge from-layer="410" from-port="0" to-layer="412" to-port="3" />
-		<edge from-layer="411" from-port="0" to-layer="412" to-port="4" />
-		<edge from-layer="412" from-port="5" to-layer="417" to-port="0" />
-		<edge from-layer="412" from-port="5" to-layer="449" to-port="0" />
-		<edge from-layer="412" from-port="5" to-layer="433" to-port="0" />
-		<edge from-layer="413" from-port="0" to-layer="414" to-port="0" />
-		<edge from-layer="414" from-port="1" to-layer="416" to-port="0" />
+		<edge from-layer="403" from-port="2" to-layer="419" to-port="1" />
+		<edge from-layer="404" from-port="0" to-layer="405" to-port="0" />
+		<edge from-layer="405" from-port="1" to-layer="408" to-port="0" />
+		<edge from-layer="406" from-port="0" to-layer="407" to-port="0" />
+		<edge from-layer="407" from-port="1" to-layer="408" to-port="1" />
+		<edge from-layer="408" from-port="2" to-layer="410" to-port="0" />
+		<edge from-layer="409" from-port="0" to-layer="410" to-port="1" />
+		<edge from-layer="410" from-port="2" to-layer="411" to-port="0" />
+		<edge from-layer="411" from-port="1" to-layer="412" to-port="1" />
+		<edge from-layer="412" from-port="2" to-layer="414" to-port="0" />
+		<edge from-layer="413" from-port="0" to-layer="414" to-port="1" />
+		<edge from-layer="414" from-port="2" to-layer="416" to-port="0" />
 		<edge from-layer="415" from-port="0" to-layer="416" to-port="1" />
-		<edge from-layer="416" from-port="2" to-layer="417" to-port="1" />
-		<edge from-layer="417" from-port="2" to-layer="419" to-port="0" />
-		<edge from-layer="418" from-port="0" to-layer="419" to-port="1" />
-		<edge from-layer="419" from-port="2" to-layer="424" to-port="0" />
-		<edge from-layer="420" from-port="0" to-layer="424" to-port="1" />
-		<edge from-layer="421" from-port="0" to-layer="424" to-port="2" />
-		<edge from-layer="422" from-port="0" to-layer="424" to-port="3" />
-		<edge from-layer="423" from-port="0" to-layer="424" to-port="4" />
-		<edge from-layer="424" from-port="5" to-layer="426" to-port="0" />
-		<edge from-layer="425" from-port="0" to-layer="426" to-port="1" />
-		<edge from-layer="426" from-port="2" to-layer="428" to-port="0" />
-		<edge from-layer="427" from-port="0" to-layer="428" to-port="1" />
-		<edge from-layer="428" from-port="2" to-layer="456" to-port="0" />
-		<edge from-layer="429" from-port="0" to-layer="430" to-port="0" />
-		<edge from-layer="430" from-port="1" to-layer="432" to-port="0" />
-		<edge from-layer="431" from-port="0" to-layer="432" to-port="1" />
-		<edge from-layer="432" from-port="2" to-layer="433" to-port="1" />
-		<edge from-layer="433" from-port="2" to-layer="435" to-port="0" />
-		<edge from-layer="434" from-port="0" to-layer="435" to-port="1" />
-		<edge from-layer="435" from-port="2" to-layer="440" to-port="0" />
-		<edge from-layer="436" from-port="0" to-layer="440" to-port="1" />
-		<edge from-layer="437" from-port="0" to-layer="440" to-port="2" />
-		<edge from-layer="438" from-port="0" to-layer="440" to-port="3" />
-		<edge from-layer="439" from-port="0" to-layer="440" to-port="4" />
-		<edge from-layer="440" from-port="5" to-layer="442" to-port="0" />
-		<edge from-layer="441" from-port="0" to-layer="442" to-port="1" />
-		<edge from-layer="442" from-port="2" to-layer="444" to-port="0" />
-		<edge from-layer="443" from-port="0" to-layer="444" to-port="1" />
-		<edge from-layer="444" from-port="2" to-layer="456" to-port="1" />
-		<edge from-layer="445" from-port="0" to-layer="446" to-port="0" />
-		<edge from-layer="446" from-port="1" to-layer="448" to-port="0" />
+		<edge from-layer="416" from-port="2" to-layer="418" to-port="0" />
+		<edge from-layer="417" from-port="0" to-layer="418" to-port="1" />
+		<edge from-layer="418" from-port="2" to-layer="419" to-port="2" />
+		<edge from-layer="419" from-port="4" to-layer="421" to-port="0" />
+		<edge from-layer="420" from-port="0" to-layer="421" to-port="1" />
+		<edge from-layer="421" from-port="2" to-layer="423" to-port="0" />
+		<edge from-layer="422" from-port="0" to-layer="423" to-port="1" />
+		<edge from-layer="423" from-port="2" to-layer="432" to-port="0" />
+		<edge from-layer="424" from-port="0" to-layer="425" to-port="0" />
+		<edge from-layer="425" from-port="1" to-layer="428" to-port="0" />
+		<edge from-layer="426" from-port="0" to-layer="427" to-port="0" />
+		<edge from-layer="427" from-port="1" to-layer="428" to-port="1" />
+		<edge from-layer="428" from-port="2" to-layer="430" to-port="0" />
+		<edge from-layer="429" from-port="0" to-layer="430" to-port="1" />
+		<edge from-layer="430" from-port="2" to-layer="431" to-port="0" />
+		<edge from-layer="431" from-port="1" to-layer="432" to-port="1" />
+		<edge from-layer="432" from-port="2" to-layer="434" to-port="0" />
+		<edge from-layer="433" from-port="0" to-layer="434" to-port="1" />
+		<edge from-layer="434" from-port="2" to-layer="435" to-port="0" />
+		<edge from-layer="435" from-port="2" to-layer="437" to-port="0" />
+		<edge from-layer="436" from-port="0" to-layer="437" to-port="1" />
+		<edge from-layer="437" from-port="2" to-layer="439" to-port="0" />
+		<edge from-layer="438" from-port="0" to-layer="439" to-port="1" />
+		<edge from-layer="439" from-port="2" to-layer="441" to-port="0" />
+		<edge from-layer="440" from-port="0" to-layer="441" to-port="1" />
+		<edge from-layer="441" from-port="2" to-layer="465" to-port="1" />
+		<edge from-layer="441" from-port="2" to-layer="450" to-port="0" />
+		<edge from-layer="442" from-port="0" to-layer="443" to-port="0" />
+		<edge from-layer="443" from-port="1" to-layer="446" to-port="0" />
+		<edge from-layer="444" from-port="0" to-layer="445" to-port="0" />
+		<edge from-layer="445" from-port="1" to-layer="446" to-port="1" />
+		<edge from-layer="446" from-port="2" to-layer="448" to-port="0" />
 		<edge from-layer="447" from-port="0" to-layer="448" to-port="1" />
-		<edge from-layer="448" from-port="2" to-layer="449" to-port="1" />
-		<edge from-layer="449" from-port="2" to-layer="451" to-port="0" />
-		<edge from-layer="450" from-port="0" to-layer="451" to-port="1" />
-		<edge from-layer="451" from-port="2" to-layer="453" to-port="0" />
-		<edge from-layer="452" from-port="0" to-layer="453" to-port="1" />
-		<edge from-layer="453" from-port="2" to-layer="455" to-port="0" />
-		<edge from-layer="454" from-port="0" to-layer="455" to-port="1" />
-		<edge from-layer="455" from-port="2" to-layer="456" to-port="2" />
-		<edge from-layer="456" from-port="4" to-layer="458" to-port="0" />
-		<edge from-layer="457" from-port="0" to-layer="458" to-port="1" />
+		<edge from-layer="448" from-port="2" to-layer="449" to-port="0" />
+		<edge from-layer="449" from-port="1" to-layer="450" to-port="1" />
+		<edge from-layer="450" from-port="2" to-layer="452" to-port="0" />
+		<edge from-layer="451" from-port="0" to-layer="452" to-port="1" />
+		<edge from-layer="452" from-port="2" to-layer="453" to-port="0" />
+		<edge from-layer="453" from-port="1" to-layer="462" to-port="0" />
+		<edge from-layer="454" from-port="0" to-layer="455" to-port="0" />
+		<edge from-layer="455" from-port="1" to-layer="458" to-port="0" />
+		<edge from-layer="456" from-port="0" to-layer="457" to-port="0" />
+		<edge from-layer="457" from-port="1" to-layer="458" to-port="1" />
 		<edge from-layer="458" from-port="2" to-layer="460" to-port="0" />
 		<edge from-layer="459" from-port="0" to-layer="460" to-port="1" />
-		<edge from-layer="460" from-port="2" to-layer="462" to-port="0" />
-		<edge from-layer="461" from-port="0" to-layer="462" to-port="1" />
-		<edge from-layer="462" from-port="2" to-layer="467" to-port="0" />
-		<edge from-layer="463" from-port="0" to-layer="467" to-port="1" />
-		<edge from-layer="464" from-port="0" to-layer="467" to-port="2" />
-		<edge from-layer="465" from-port="0" to-layer="467" to-port="3" />
-		<edge from-layer="466" from-port="0" to-layer="467" to-port="4" />
-		<edge from-layer="467" from-port="5" to-layer="472" to-port="0" />
-		<edge from-layer="468" from-port="0" to-layer="469" to-port="0" />
-		<edge from-layer="469" from-port="1" to-layer="471" to-port="0" />
+		<edge from-layer="460" from-port="2" to-layer="461" to-port="0" />
+		<edge from-layer="461" from-port="1" to-layer="462" to-port="1" />
+		<edge from-layer="462" from-port="2" to-layer="464" to-port="0" />
+		<edge from-layer="463" from-port="0" to-layer="464" to-port="1" />
+		<edge from-layer="464" from-port="2" to-layer="465" to-port="0" />
+		<edge from-layer="465" from-port="2" to-layer="467" to-port="0" />
+		<edge from-layer="466" from-port="0" to-layer="467" to-port="1" />
+		<edge from-layer="467" from-port="2" to-layer="469" to-port="0" />
+		<edge from-layer="468" from-port="0" to-layer="469" to-port="1" />
+		<edge from-layer="469" from-port="2" to-layer="471" to-port="0" />
 		<edge from-layer="470" from-port="0" to-layer="471" to-port="1" />
-		<edge from-layer="471" from-port="2" to-layer="472" to-port="1" />
-		<edge from-layer="472" from-port="2" to-layer="474" to-port="0" />
-		<edge from-layer="473" from-port="0" to-layer="474" to-port="1" />
-		<edge from-layer="474" from-port="2" to-layer="475" to-port="0" />
-		<edge from-layer="475" from-port="2" to-layer="477" to-port="0" />
-		<edge from-layer="476" from-port="0" to-layer="477" to-port="1" />
-		<edge from-layer="477" from-port="2" to-layer="479" to-port="0" />
-		<edge from-layer="478" from-port="0" to-layer="479" to-port="1" />
-		<edge from-layer="479" from-port="2" to-layer="481" to-port="0" />
-		<edge from-layer="480" from-port="0" to-layer="481" to-port="1" />
-		<edge from-layer="481" from-port="2" to-layer="483" to-port="0" />
-		<edge from-layer="481" from-port="2" to-layer="511" to-port="1" />
-		<edge from-layer="482" from-port="0" to-layer="483" to-port="1" />
-		<edge from-layer="483" from-port="2" to-layer="488" to-port="0" />
-		<edge from-layer="484" from-port="0" to-layer="488" to-port="1" />
-		<edge from-layer="485" from-port="0" to-layer="488" to-port="2" />
-		<edge from-layer="486" from-port="0" to-layer="488" to-port="3" />
-		<edge from-layer="487" from-port="0" to-layer="488" to-port="4" />
-		<edge from-layer="488" from-port="5" to-layer="493" to-port="0" />
+		<edge from-layer="471" from-port="2" to-layer="510" to-port="0" />
+		<edge from-layer="471" from-port="2" to-layer="533" to-port="1" />
+		<edge from-layer="471" from-port="2" to-layer="480" to-port="0" />
+		<edge from-layer="471" from-port="2" to-layer="495" to-port="0" />
+		<edge from-layer="472" from-port="0" to-layer="473" to-port="0" />
+		<edge from-layer="473" from-port="1" to-layer="476" to-port="0" />
+		<edge from-layer="474" from-port="0" to-layer="475" to-port="0" />
+		<edge from-layer="475" from-port="1" to-layer="476" to-port="1" />
+		<edge from-layer="476" from-port="2" to-layer="478" to-port="0" />
+		<edge from-layer="477" from-port="0" to-layer="478" to-port="1" />
+		<edge from-layer="478" from-port="2" to-layer="479" to-port="0" />
+		<edge from-layer="479" from-port="1" to-layer="480" to-port="1" />
+		<edge from-layer="480" from-port="2" to-layer="482" to-port="0" />
+		<edge from-layer="481" from-port="0" to-layer="482" to-port="1" />
+		<edge from-layer="482" from-port="2" to-layer="484" to-port="0" />
+		<edge from-layer="483" from-port="0" to-layer="484" to-port="1" />
+		<edge from-layer="484" from-port="2" to-layer="486" to-port="0" />
+		<edge from-layer="485" from-port="0" to-layer="486" to-port="1" />
+		<edge from-layer="486" from-port="2" to-layer="517" to-port="0" />
+		<edge from-layer="487" from-port="0" to-layer="488" to-port="0" />
+		<edge from-layer="488" from-port="1" to-layer="491" to-port="0" />
 		<edge from-layer="489" from-port="0" to-layer="490" to-port="0" />
-		<edge from-layer="490" from-port="1" to-layer="492" to-port="0" />
-		<edge from-layer="491" from-port="0" to-layer="492" to-port="1" />
-		<edge from-layer="492" from-port="2" to-layer="493" to-port="1" />
-		<edge from-layer="493" from-port="2" to-layer="495" to-port="0" />
-		<edge from-layer="494" from-port="0" to-layer="495" to-port="1" />
-		<edge from-layer="495" from-port="2" to-layer="496" to-port="0" />
-		<edge from-layer="496" from-port="1" to-layer="498" to-port="0" />
-		<edge from-layer="497" from-port="0" to-layer="498" to-port="1" />
-		<edge from-layer="498" from-port="2" to-layer="503" to-port="0" />
-		<edge from-layer="499" from-port="0" to-layer="503" to-port="1" />
-		<edge from-layer="500" from-port="0" to-layer="503" to-port="2" />
-		<edge from-layer="501" from-port="0" to-layer="503" to-port="3" />
-		<edge from-layer="502" from-port="0" to-layer="503" to-port="4" />
-		<edge from-layer="503" from-port="5" to-layer="508" to-port="0" />
+		<edge from-layer="490" from-port="1" to-layer="491" to-port="1" />
+		<edge from-layer="491" from-port="2" to-layer="493" to-port="0" />
+		<edge from-layer="492" from-port="0" to-layer="493" to-port="1" />
+		<edge from-layer="493" from-port="2" to-layer="494" to-port="0" />
+		<edge from-layer="494" from-port="1" to-layer="495" to-port="1" />
+		<edge from-layer="495" from-port="2" to-layer="497" to-port="0" />
+		<edge from-layer="496" from-port="0" to-layer="497" to-port="1" />
+		<edge from-layer="497" from-port="2" to-layer="499" to-port="0" />
+		<edge from-layer="498" from-port="0" to-layer="499" to-port="1" />
+		<edge from-layer="499" from-port="2" to-layer="501" to-port="0" />
+		<edge from-layer="500" from-port="0" to-layer="501" to-port="1" />
+		<edge from-layer="501" from-port="2" to-layer="517" to-port="1" />
+		<edge from-layer="502" from-port="0" to-layer="503" to-port="0" />
+		<edge from-layer="503" from-port="1" to-layer="506" to-port="0" />
 		<edge from-layer="504" from-port="0" to-layer="505" to-port="0" />
-		<edge from-layer="505" from-port="1" to-layer="507" to-port="0" />
-		<edge from-layer="506" from-port="0" to-layer="507" to-port="1" />
-		<edge from-layer="507" from-port="2" to-layer="508" to-port="1" />
-		<edge from-layer="508" from-port="2" to-layer="510" to-port="0" />
-		<edge from-layer="509" from-port="0" to-layer="510" to-port="1" />
-		<edge from-layer="510" from-port="2" to-layer="511" to-port="0" />
-		<edge from-layer="511" from-port="2" to-layer="513" to-port="0" />
-		<edge from-layer="512" from-port="0" to-layer="513" to-port="1" />
-		<edge from-layer="513" from-port="2" to-layer="515" to-port="0" />
-		<edge from-layer="514" from-port="0" to-layer="515" to-port="1" />
-		<edge from-layer="515" from-port="2" to-layer="517" to-port="0" />
-		<edge from-layer="516" from-port="0" to-layer="517" to-port="1" />
-		<edge from-layer="517" from-port="2" to-layer="519" to-port="0" />
-		<edge from-layer="517" from-port="2" to-layer="587" to-port="1" />
+		<edge from-layer="505" from-port="1" to-layer="506" to-port="1" />
+		<edge from-layer="506" from-port="2" to-layer="508" to-port="0" />
+		<edge from-layer="507" from-port="0" to-layer="508" to-port="1" />
+		<edge from-layer="508" from-port="2" to-layer="509" to-port="0" />
+		<edge from-layer="509" from-port="1" to-layer="510" to-port="1" />
+		<edge from-layer="510" from-port="2" to-layer="512" to-port="0" />
+		<edge from-layer="511" from-port="0" to-layer="512" to-port="1" />
+		<edge from-layer="512" from-port="2" to-layer="514" to-port="0" />
+		<edge from-layer="513" from-port="0" to-layer="514" to-port="1" />
+		<edge from-layer="514" from-port="2" to-layer="516" to-port="0" />
+		<edge from-layer="515" from-port="0" to-layer="516" to-port="1" />
+		<edge from-layer="516" from-port="2" to-layer="517" to-port="2" />
+		<edge from-layer="517" from-port="4" to-layer="519" to-port="0" />
 		<edge from-layer="518" from-port="0" to-layer="519" to-port="1" />
-		<edge from-layer="519" from-port="2" to-layer="524" to-port="0" />
-		<edge from-layer="520" from-port="0" to-layer="524" to-port="1" />
-		<edge from-layer="521" from-port="0" to-layer="524" to-port="2" />
-		<edge from-layer="522" from-port="0" to-layer="524" to-port="3" />
-		<edge from-layer="523" from-port="0" to-layer="524" to-port="4" />
-		<edge from-layer="524" from-port="5" to-layer="529" to-port="0" />
-		<edge from-layer="524" from-port="5" to-layer="545" to-port="0" />
-		<edge from-layer="524" from-port="5" to-layer="561" to-port="0" />
-		<edge from-layer="525" from-port="0" to-layer="526" to-port="0" />
-		<edge from-layer="526" from-port="1" to-layer="528" to-port="0" />
+		<edge from-layer="519" from-port="2" to-layer="521" to-port="0" />
+		<edge from-layer="520" from-port="0" to-layer="521" to-port="1" />
+		<edge from-layer="521" from-port="2" to-layer="530" to-port="0" />
+		<edge from-layer="522" from-port="0" to-layer="523" to-port="0" />
+		<edge from-layer="523" from-port="1" to-layer="526" to-port="0" />
+		<edge from-layer="524" from-port="0" to-layer="525" to-port="0" />
+		<edge from-layer="525" from-port="1" to-layer="526" to-port="1" />
+		<edge from-layer="526" from-port="2" to-layer="528" to-port="0" />
 		<edge from-layer="527" from-port="0" to-layer="528" to-port="1" />
-		<edge from-layer="528" from-port="2" to-layer="529" to-port="1" />
-		<edge from-layer="529" from-port="2" to-layer="531" to-port="0" />
-		<edge from-layer="530" from-port="0" to-layer="531" to-port="1" />
-		<edge from-layer="531" from-port="2" to-layer="536" to-port="0" />
-		<edge from-layer="532" from-port="0" to-layer="536" to-port="1" />
-		<edge from-layer="533" from-port="0" to-layer="536" to-port="2" />
-		<edge from-layer="534" from-port="0" to-layer="536" to-port="3" />
-		<edge from-layer="535" from-port="0" to-layer="536" to-port="4" />
-		<edge from-layer="536" from-port="5" to-layer="538" to-port="0" />
-		<edge from-layer="537" from-port="0" to-layer="538" to-port="1" />
-		<edge from-layer="538" from-port="2" to-layer="540" to-port="0" />
-		<edge from-layer="539" from-port="0" to-layer="540" to-port="1" />
-		<edge from-layer="540" from-port="2" to-layer="568" to-port="0" />
-		<edge from-layer="541" from-port="0" to-layer="542" to-port="0" />
-		<edge from-layer="542" from-port="1" to-layer="544" to-port="0" />
-		<edge from-layer="543" from-port="0" to-layer="544" to-port="1" />
-		<edge from-layer="544" from-port="2" to-layer="545" to-port="1" />
-		<edge from-layer="545" from-port="2" to-layer="547" to-port="0" />
-		<edge from-layer="546" from-port="0" to-layer="547" to-port="1" />
-		<edge from-layer="547" from-port="2" to-layer="552" to-port="0" />
-		<edge from-layer="548" from-port="0" to-layer="552" to-port="1" />
-		<edge from-layer="549" from-port="0" to-layer="552" to-port="2" />
-		<edge from-layer="550" from-port="0" to-layer="552" to-port="3" />
-		<edge from-layer="551" from-port="0" to-layer="552" to-port="4" />
-		<edge from-layer="552" from-port="5" to-layer="554" to-port="0" />
-		<edge from-layer="553" from-port="0" to-layer="554" to-port="1" />
-		<edge from-layer="554" from-port="2" to-layer="556" to-port="0" />
-		<edge from-layer="555" from-port="0" to-layer="556" to-port="1" />
-		<edge from-layer="556" from-port="2" to-layer="568" to-port="1" />
-		<edge from-layer="557" from-port="0" to-layer="558" to-port="0" />
-		<edge from-layer="558" from-port="1" to-layer="560" to-port="0" />
-		<edge from-layer="559" from-port="0" to-layer="560" to-port="1" />
-		<edge from-layer="560" from-port="2" to-layer="561" to-port="1" />
-		<edge from-layer="561" from-port="2" to-layer="563" to-port="0" />
-		<edge from-layer="562" from-port="0" to-layer="563" to-port="1" />
+		<edge from-layer="528" from-port="2" to-layer="529" to-port="0" />
+		<edge from-layer="529" from-port="1" to-layer="530" to-port="1" />
+		<edge from-layer="530" from-port="2" to-layer="532" to-port="0" />
+		<edge from-layer="531" from-port="0" to-layer="532" to-port="1" />
+		<edge from-layer="532" from-port="2" to-layer="533" to-port="0" />
+		<edge from-layer="533" from-port="2" to-layer="535" to-port="0" />
+		<edge from-layer="534" from-port="0" to-layer="535" to-port="1" />
+		<edge from-layer="535" from-port="2" to-layer="537" to-port="0" />
+		<edge from-layer="536" from-port="0" to-layer="537" to-port="1" />
+		<edge from-layer="537" from-port="2" to-layer="539" to-port="0" />
+		<edge from-layer="538" from-port="0" to-layer="539" to-port="1" />
+		<edge from-layer="539" from-port="2" to-layer="563" to-port="1" />
+		<edge from-layer="539" from-port="2" to-layer="548" to-port="0" />
+		<edge from-layer="540" from-port="0" to-layer="541" to-port="0" />
+		<edge from-layer="541" from-port="1" to-layer="544" to-port="0" />
+		<edge from-layer="542" from-port="0" to-layer="543" to-port="0" />
+		<edge from-layer="543" from-port="1" to-layer="544" to-port="1" />
+		<edge from-layer="544" from-port="2" to-layer="546" to-port="0" />
+		<edge from-layer="545" from-port="0" to-layer="546" to-port="1" />
+		<edge from-layer="546" from-port="2" to-layer="547" to-port="0" />
+		<edge from-layer="547" from-port="1" to-layer="548" to-port="1" />
+		<edge from-layer="548" from-port="2" to-layer="550" to-port="0" />
+		<edge from-layer="549" from-port="0" to-layer="550" to-port="1" />
+		<edge from-layer="550" from-port="2" to-layer="551" to-port="0" />
+		<edge from-layer="551" from-port="1" to-layer="560" to-port="0" />
+		<edge from-layer="552" from-port="0" to-layer="553" to-port="0" />
+		<edge from-layer="553" from-port="1" to-layer="556" to-port="0" />
+		<edge from-layer="554" from-port="0" to-layer="555" to-port="0" />
+		<edge from-layer="555" from-port="1" to-layer="556" to-port="1" />
+		<edge from-layer="556" from-port="2" to-layer="558" to-port="0" />
+		<edge from-layer="557" from-port="0" to-layer="558" to-port="1" />
+		<edge from-layer="558" from-port="2" to-layer="559" to-port="0" />
+		<edge from-layer="559" from-port="1" to-layer="560" to-port="1" />
+		<edge from-layer="560" from-port="2" to-layer="562" to-port="0" />
+		<edge from-layer="561" from-port="0" to-layer="562" to-port="1" />
+		<edge from-layer="562" from-port="2" to-layer="563" to-port="0" />
 		<edge from-layer="563" from-port="2" to-layer="565" to-port="0" />
 		<edge from-layer="564" from-port="0" to-layer="565" to-port="1" />
 		<edge from-layer="565" from-port="2" to-layer="567" to-port="0" />
 		<edge from-layer="566" from-port="0" to-layer="567" to-port="1" />
-		<edge from-layer="567" from-port="2" to-layer="568" to-port="2" />
-		<edge from-layer="568" from-port="4" to-layer="570" to-port="0" />
-		<edge from-layer="569" from-port="0" to-layer="570" to-port="1" />
-		<edge from-layer="570" from-port="2" to-layer="572" to-port="0" />
-		<edge from-layer="571" from-port="0" to-layer="572" to-port="1" />
-		<edge from-layer="572" from-port="2" to-layer="574" to-port="0" />
-		<edge from-layer="573" from-port="0" to-layer="574" to-port="1" />
-		<edge from-layer="574" from-port="2" to-layer="579" to-port="0" />
-		<edge from-layer="575" from-port="0" to-layer="579" to-port="1" />
-		<edge from-layer="576" from-port="0" to-layer="579" to-port="2" />
-		<edge from-layer="577" from-port="0" to-layer="579" to-port="3" />
-		<edge from-layer="578" from-port="0" to-layer="579" to-port="4" />
-		<edge from-layer="579" from-port="5" to-layer="584" to-port="0" />
-		<edge from-layer="580" from-port="0" to-layer="581" to-port="0" />
-		<edge from-layer="581" from-port="1" to-layer="583" to-port="0" />
-		<edge from-layer="582" from-port="0" to-layer="583" to-port="1" />
-		<edge from-layer="583" from-port="2" to-layer="584" to-port="1" />
-		<edge from-layer="584" from-port="2" to-layer="586" to-port="0" />
-		<edge from-layer="585" from-port="0" to-layer="586" to-port="1" />
-		<edge from-layer="586" from-port="2" to-layer="587" to-port="0" />
-		<edge from-layer="587" from-port="2" to-layer="589" to-port="0" />
-		<edge from-layer="588" from-port="0" to-layer="589" to-port="1" />
+		<edge from-layer="567" from-port="2" to-layer="569" to-port="0" />
+		<edge from-layer="568" from-port="0" to-layer="569" to-port="1" />
+		<edge from-layer="569" from-port="2" to-layer="608" to-port="0" />
+		<edge from-layer="569" from-port="2" to-layer="631" to-port="1" />
+		<edge from-layer="569" from-port="2" to-layer="593" to-port="0" />
+		<edge from-layer="569" from-port="2" to-layer="578" to-port="0" />
+		<edge from-layer="570" from-port="0" to-layer="571" to-port="0" />
+		<edge from-layer="571" from-port="1" to-layer="574" to-port="0" />
+		<edge from-layer="572" from-port="0" to-layer="573" to-port="0" />
+		<edge from-layer="573" from-port="1" to-layer="574" to-port="1" />
+		<edge from-layer="574" from-port="2" to-layer="576" to-port="0" />
+		<edge from-layer="575" from-port="0" to-layer="576" to-port="1" />
+		<edge from-layer="576" from-port="2" to-layer="577" to-port="0" />
+		<edge from-layer="577" from-port="1" to-layer="578" to-port="1" />
+		<edge from-layer="578" from-port="2" to-layer="580" to-port="0" />
+		<edge from-layer="579" from-port="0" to-layer="580" to-port="1" />
+		<edge from-layer="580" from-port="2" to-layer="582" to-port="0" />
+		<edge from-layer="581" from-port="0" to-layer="582" to-port="1" />
+		<edge from-layer="582" from-port="2" to-layer="584" to-port="0" />
+		<edge from-layer="583" from-port="0" to-layer="584" to-port="1" />
+		<edge from-layer="584" from-port="2" to-layer="615" to-port="0" />
+		<edge from-layer="585" from-port="0" to-layer="586" to-port="0" />
+		<edge from-layer="586" from-port="1" to-layer="589" to-port="0" />
+		<edge from-layer="587" from-port="0" to-layer="588" to-port="0" />
+		<edge from-layer="588" from-port="1" to-layer="589" to-port="1" />
 		<edge from-layer="589" from-port="2" to-layer="591" to-port="0" />
 		<edge from-layer="590" from-port="0" to-layer="591" to-port="1" />
-		<edge from-layer="591" from-port="2" to-layer="593" to-port="0" />
-		<edge from-layer="592" from-port="0" to-layer="593" to-port="1" />
-		<edge from-layer="593" from-port="2" to-layer="623" to-port="1" />
+		<edge from-layer="591" from-port="2" to-layer="592" to-port="0" />
+		<edge from-layer="592" from-port="1" to-layer="593" to-port="1" />
 		<edge from-layer="593" from-port="2" to-layer="595" to-port="0" />
 		<edge from-layer="594" from-port="0" to-layer="595" to-port="1" />
-		<edge from-layer="595" from-port="2" to-layer="600" to-port="0" />
-		<edge from-layer="596" from-port="0" to-layer="600" to-port="1" />
-		<edge from-layer="597" from-port="0" to-layer="600" to-port="2" />
-		<edge from-layer="598" from-port="0" to-layer="600" to-port="3" />
-		<edge from-layer="599" from-port="0" to-layer="600" to-port="4" />
-		<edge from-layer="600" from-port="5" to-layer="605" to-port="0" />
-		<edge from-layer="601" from-port="0" to-layer="602" to-port="0" />
-		<edge from-layer="602" from-port="1" to-layer="604" to-port="0" />
-		<edge from-layer="603" from-port="0" to-layer="604" to-port="1" />
-		<edge from-layer="604" from-port="2" to-layer="605" to-port="1" />
-		<edge from-layer="605" from-port="2" to-layer="607" to-port="0" />
-		<edge from-layer="606" from-port="0" to-layer="607" to-port="1" />
-		<edge from-layer="607" from-port="2" to-layer="608" to-port="0" />
-		<edge from-layer="608" from-port="1" to-layer="610" to-port="0" />
+		<edge from-layer="595" from-port="2" to-layer="597" to-port="0" />
+		<edge from-layer="596" from-port="0" to-layer="597" to-port="1" />
+		<edge from-layer="597" from-port="2" to-layer="599" to-port="0" />
+		<edge from-layer="598" from-port="0" to-layer="599" to-port="1" />
+		<edge from-layer="599" from-port="2" to-layer="615" to-port="1" />
+		<edge from-layer="600" from-port="0" to-layer="601" to-port="0" />
+		<edge from-layer="601" from-port="1" to-layer="604" to-port="0" />
+		<edge from-layer="602" from-port="0" to-layer="603" to-port="0" />
+		<edge from-layer="603" from-port="1" to-layer="604" to-port="1" />
+		<edge from-layer="604" from-port="2" to-layer="606" to-port="0" />
+		<edge from-layer="605" from-port="0" to-layer="606" to-port="1" />
+		<edge from-layer="606" from-port="2" to-layer="607" to-port="0" />
+		<edge from-layer="607" from-port="1" to-layer="608" to-port="1" />
+		<edge from-layer="608" from-port="2" to-layer="610" to-port="0" />
 		<edge from-layer="609" from-port="0" to-layer="610" to-port="1" />
-		<edge from-layer="610" from-port="2" to-layer="615" to-port="0" />
-		<edge from-layer="611" from-port="0" to-layer="615" to-port="1" />
-		<edge from-layer="612" from-port="0" to-layer="615" to-port="2" />
-		<edge from-layer="613" from-port="0" to-layer="615" to-port="3" />
-		<edge from-layer="614" from-port="0" to-layer="615" to-port="4" />
-		<edge from-layer="615" from-port="5" to-layer="620" to-port="0" />
-		<edge from-layer="616" from-port="0" to-layer="617" to-port="0" />
-		<edge from-layer="617" from-port="1" to-layer="619" to-port="0" />
+		<edge from-layer="610" from-port="2" to-layer="612" to-port="0" />
+		<edge from-layer="611" from-port="0" to-layer="612" to-port="1" />
+		<edge from-layer="612" from-port="2" to-layer="614" to-port="0" />
+		<edge from-layer="613" from-port="0" to-layer="614" to-port="1" />
+		<edge from-layer="614" from-port="2" to-layer="615" to-port="2" />
+		<edge from-layer="615" from-port="4" to-layer="617" to-port="0" />
+		<edge from-layer="616" from-port="0" to-layer="617" to-port="1" />
+		<edge from-layer="617" from-port="2" to-layer="619" to-port="0" />
 		<edge from-layer="618" from-port="0" to-layer="619" to-port="1" />
-		<edge from-layer="619" from-port="2" to-layer="620" to-port="1" />
-		<edge from-layer="620" from-port="2" to-layer="622" to-port="0" />
-		<edge from-layer="621" from-port="0" to-layer="622" to-port="1" />
-		<edge from-layer="622" from-port="2" to-layer="623" to-port="0" />
-		<edge from-layer="623" from-port="2" to-layer="625" to-port="0" />
-		<edge from-layer="624" from-port="0" to-layer="625" to-port="1" />
-		<edge from-layer="625" from-port="2" to-layer="627" to-port="0" />
-		<edge from-layer="626" from-port="0" to-layer="627" to-port="1" />
-		<edge from-layer="627" from-port="2" to-layer="629" to-port="0" />
-		<edge from-layer="628" from-port="0" to-layer="629" to-port="1" />
-		<edge from-layer="629" from-port="2" to-layer="699" to-port="1" />
-		<edge from-layer="629" from-port="2" to-layer="631" to-port="0" />
-		<edge from-layer="630" from-port="0" to-layer="631" to-port="1" />
-		<edge from-layer="631" from-port="2" to-layer="636" to-port="0" />
-		<edge from-layer="632" from-port="0" to-layer="636" to-port="1" />
-		<edge from-layer="633" from-port="0" to-layer="636" to-port="2" />
-		<edge from-layer="634" from-port="0" to-layer="636" to-port="3" />
-		<edge from-layer="635" from-port="0" to-layer="636" to-port="4" />
-		<edge from-layer="636" from-port="5" to-layer="673" to-port="0" />
-		<edge from-layer="636" from-port="5" to-layer="657" to-port="0" />
-		<edge from-layer="636" from-port="5" to-layer="641" to-port="0" />
-		<edge from-layer="637" from-port="0" to-layer="638" to-port="0" />
-		<edge from-layer="638" from-port="1" to-layer="640" to-port="0" />
-		<edge from-layer="639" from-port="0" to-layer="640" to-port="1" />
-		<edge from-layer="640" from-port="2" to-layer="641" to-port="1" />
-		<edge from-layer="641" from-port="2" to-layer="643" to-port="0" />
-		<edge from-layer="642" from-port="0" to-layer="643" to-port="1" />
-		<edge from-layer="643" from-port="2" to-layer="648" to-port="0" />
-		<edge from-layer="644" from-port="0" to-layer="648" to-port="1" />
-		<edge from-layer="645" from-port="0" to-layer="648" to-port="2" />
-		<edge from-layer="646" from-port="0" to-layer="648" to-port="3" />
-		<edge from-layer="647" from-port="0" to-layer="648" to-port="4" />
-		<edge from-layer="648" from-port="5" to-layer="650" to-port="0" />
-		<edge from-layer="649" from-port="0" to-layer="650" to-port="1" />
-		<edge from-layer="650" from-port="2" to-layer="652" to-port="0" />
-		<edge from-layer="651" from-port="0" to-layer="652" to-port="1" />
-		<edge from-layer="652" from-port="2" to-layer="680" to-port="0" />
-		<edge from-layer="653" from-port="0" to-layer="654" to-port="0" />
-		<edge from-layer="654" from-port="1" to-layer="656" to-port="0" />
+		<edge from-layer="619" from-port="2" to-layer="628" to-port="0" />
+		<edge from-layer="620" from-port="0" to-layer="621" to-port="0" />
+		<edge from-layer="621" from-port="1" to-layer="624" to-port="0" />
+		<edge from-layer="622" from-port="0" to-layer="623" to-port="0" />
+		<edge from-layer="623" from-port="1" to-layer="624" to-port="1" />
+		<edge from-layer="624" from-port="2" to-layer="626" to-port="0" />
+		<edge from-layer="625" from-port="0" to-layer="626" to-port="1" />
+		<edge from-layer="626" from-port="2" to-layer="627" to-port="0" />
+		<edge from-layer="627" from-port="1" to-layer="628" to-port="1" />
+		<edge from-layer="628" from-port="2" to-layer="630" to-port="0" />
+		<edge from-layer="629" from-port="0" to-layer="630" to-port="1" />
+		<edge from-layer="630" from-port="2" to-layer="631" to-port="0" />
+		<edge from-layer="631" from-port="2" to-layer="633" to-port="0" />
+		<edge from-layer="632" from-port="0" to-layer="633" to-port="1" />
+		<edge from-layer="633" from-port="2" to-layer="635" to-port="0" />
+		<edge from-layer="634" from-port="0" to-layer="635" to-port="1" />
+		<edge from-layer="635" from-port="2" to-layer="637" to-port="0" />
+		<edge from-layer="636" from-port="0" to-layer="637" to-port="1" />
+		<edge from-layer="637" from-port="2" to-layer="646" to-port="0" />
+		<edge from-layer="637" from-port="2" to-layer="661" to-port="1" />
+		<edge from-layer="638" from-port="0" to-layer="639" to-port="0" />
+		<edge from-layer="639" from-port="1" to-layer="642" to-port="0" />
+		<edge from-layer="640" from-port="0" to-layer="641" to-port="0" />
+		<edge from-layer="641" from-port="1" to-layer="642" to-port="1" />
+		<edge from-layer="642" from-port="2" to-layer="644" to-port="0" />
+		<edge from-layer="643" from-port="0" to-layer="644" to-port="1" />
+		<edge from-layer="644" from-port="2" to-layer="645" to-port="0" />
+		<edge from-layer="645" from-port="1" to-layer="646" to-port="1" />
+		<edge from-layer="646" from-port="2" to-layer="648" to-port="0" />
+		<edge from-layer="647" from-port="0" to-layer="648" to-port="1" />
+		<edge from-layer="648" from-port="2" to-layer="649" to-port="0" />
+		<edge from-layer="649" from-port="1" to-layer="658" to-port="0" />
+		<edge from-layer="650" from-port="0" to-layer="651" to-port="0" />
+		<edge from-layer="651" from-port="1" to-layer="654" to-port="0" />
+		<edge from-layer="652" from-port="0" to-layer="653" to-port="0" />
+		<edge from-layer="653" from-port="1" to-layer="654" to-port="1" />
+		<edge from-layer="654" from-port="2" to-layer="656" to-port="0" />
 		<edge from-layer="655" from-port="0" to-layer="656" to-port="1" />
-		<edge from-layer="656" from-port="2" to-layer="657" to-port="1" />
-		<edge from-layer="657" from-port="2" to-layer="659" to-port="0" />
-		<edge from-layer="658" from-port="0" to-layer="659" to-port="1" />
-		<edge from-layer="659" from-port="2" to-layer="664" to-port="0" />
-		<edge from-layer="660" from-port="0" to-layer="664" to-port="1" />
-		<edge from-layer="661" from-port="0" to-layer="664" to-port="2" />
-		<edge from-layer="662" from-port="0" to-layer="664" to-port="3" />
-		<edge from-layer="663" from-port="0" to-layer="664" to-port="4" />
-		<edge from-layer="664" from-port="5" to-layer="666" to-port="0" />
-		<edge from-layer="665" from-port="0" to-layer="666" to-port="1" />
-		<edge from-layer="666" from-port="2" to-layer="668" to-port="0" />
-		<edge from-layer="667" from-port="0" to-layer="668" to-port="1" />
-		<edge from-layer="668" from-port="2" to-layer="680" to-port="1" />
-		<edge from-layer="669" from-port="0" to-layer="670" to-port="0" />
-		<edge from-layer="670" from-port="1" to-layer="672" to-port="0" />
-		<edge from-layer="671" from-port="0" to-layer="672" to-port="1" />
-		<edge from-layer="672" from-port="2" to-layer="673" to-port="1" />
-		<edge from-layer="673" from-port="2" to-layer="675" to-port="0" />
-		<edge from-layer="674" from-port="0" to-layer="675" to-port="1" />
-		<edge from-layer="675" from-port="2" to-layer="677" to-port="0" />
-		<edge from-layer="676" from-port="0" to-layer="677" to-port="1" />
-		<edge from-layer="677" from-port="2" to-layer="679" to-port="0" />
-		<edge from-layer="678" from-port="0" to-layer="679" to-port="1" />
-		<edge from-layer="679" from-port="2" to-layer="680" to-port="2" />
-		<edge from-layer="680" from-port="4" to-layer="682" to-port="0" />
+		<edge from-layer="656" from-port="2" to-layer="657" to-port="0" />
+		<edge from-layer="657" from-port="1" to-layer="658" to-port="1" />
+		<edge from-layer="658" from-port="2" to-layer="660" to-port="0" />
+		<edge from-layer="659" from-port="0" to-layer="660" to-port="1" />
+		<edge from-layer="660" from-port="2" to-layer="661" to-port="0" />
+		<edge from-layer="661" from-port="2" to-layer="663" to-port="0" />
+		<edge from-layer="662" from-port="0" to-layer="663" to-port="1" />
+		<edge from-layer="663" from-port="2" to-layer="665" to-port="0" />
+		<edge from-layer="664" from-port="0" to-layer="665" to-port="1" />
+		<edge from-layer="665" from-port="2" to-layer="667" to-port="0" />
+		<edge from-layer="666" from-port="0" to-layer="667" to-port="1" />
+		<edge from-layer="667" from-port="2" to-layer="706" to-port="0" />
+		<edge from-layer="667" from-port="2" to-layer="729" to-port="1" />
+		<edge from-layer="667" from-port="2" to-layer="676" to-port="0" />
+		<edge from-layer="667" from-port="2" to-layer="691" to-port="0" />
+		<edge from-layer="668" from-port="0" to-layer="669" to-port="0" />
+		<edge from-layer="669" from-port="1" to-layer="672" to-port="0" />
+		<edge from-layer="670" from-port="0" to-layer="671" to-port="0" />
+		<edge from-layer="671" from-port="1" to-layer="672" to-port="1" />
+		<edge from-layer="672" from-port="2" to-layer="674" to-port="0" />
+		<edge from-layer="673" from-port="0" to-layer="674" to-port="1" />
+		<edge from-layer="674" from-port="2" to-layer="675" to-port="0" />
+		<edge from-layer="675" from-port="1" to-layer="676" to-port="1" />
+		<edge from-layer="676" from-port="2" to-layer="678" to-port="0" />
+		<edge from-layer="677" from-port="0" to-layer="678" to-port="1" />
+		<edge from-layer="678" from-port="2" to-layer="680" to-port="0" />
+		<edge from-layer="679" from-port="0" to-layer="680" to-port="1" />
+		<edge from-layer="680" from-port="2" to-layer="682" to-port="0" />
 		<edge from-layer="681" from-port="0" to-layer="682" to-port="1" />
-		<edge from-layer="682" from-port="2" to-layer="684" to-port="0" />
-		<edge from-layer="683" from-port="0" to-layer="684" to-port="1" />
-		<edge from-layer="684" from-port="2" to-layer="686" to-port="0" />
-		<edge from-layer="685" from-port="0" to-layer="686" to-port="1" />
-		<edge from-layer="686" from-port="2" to-layer="691" to-port="0" />
-		<edge from-layer="687" from-port="0" to-layer="691" to-port="1" />
-		<edge from-layer="688" from-port="0" to-layer="691" to-port="2" />
-		<edge from-layer="689" from-port="0" to-layer="691" to-port="3" />
-		<edge from-layer="690" from-port="0" to-layer="691" to-port="4" />
-		<edge from-layer="691" from-port="5" to-layer="696" to-port="0" />
-		<edge from-layer="692" from-port="0" to-layer="693" to-port="0" />
-		<edge from-layer="693" from-port="1" to-layer="695" to-port="0" />
+		<edge from-layer="682" from-port="2" to-layer="713" to-port="0" />
+		<edge from-layer="683" from-port="0" to-layer="684" to-port="0" />
+		<edge from-layer="684" from-port="1" to-layer="687" to-port="0" />
+		<edge from-layer="685" from-port="0" to-layer="686" to-port="0" />
+		<edge from-layer="686" from-port="1" to-layer="687" to-port="1" />
+		<edge from-layer="687" from-port="2" to-layer="689" to-port="0" />
+		<edge from-layer="688" from-port="0" to-layer="689" to-port="1" />
+		<edge from-layer="689" from-port="2" to-layer="690" to-port="0" />
+		<edge from-layer="690" from-port="1" to-layer="691" to-port="1" />
+		<edge from-layer="691" from-port="2" to-layer="693" to-port="0" />
+		<edge from-layer="692" from-port="0" to-layer="693" to-port="1" />
+		<edge from-layer="693" from-port="2" to-layer="695" to-port="0" />
 		<edge from-layer="694" from-port="0" to-layer="695" to-port="1" />
-		<edge from-layer="695" from-port="2" to-layer="696" to-port="1" />
-		<edge from-layer="696" from-port="2" to-layer="698" to-port="0" />
-		<edge from-layer="697" from-port="0" to-layer="698" to-port="1" />
-		<edge from-layer="698" from-port="2" to-layer="699" to-port="0" />
-		<edge from-layer="699" from-port="2" to-layer="701" to-port="0" />
-		<edge from-layer="700" from-port="0" to-layer="701" to-port="1" />
-		<edge from-layer="701" from-port="2" to-layer="703" to-port="0" />
-		<edge from-layer="702" from-port="0" to-layer="703" to-port="1" />
-		<edge from-layer="703" from-port="2" to-layer="705" to-port="0" />
-		<edge from-layer="704" from-port="0" to-layer="705" to-port="1" />
-		<edge from-layer="705" from-port="2" to-layer="735" to-port="1" />
-		<edge from-layer="705" from-port="2" to-layer="707" to-port="0" />
-		<edge from-layer="706" from-port="0" to-layer="707" to-port="1" />
-		<edge from-layer="707" from-port="2" to-layer="712" to-port="0" />
-		<edge from-layer="708" from-port="0" to-layer="712" to-port="1" />
-		<edge from-layer="709" from-port="0" to-layer="712" to-port="2" />
-		<edge from-layer="710" from-port="0" to-layer="712" to-port="3" />
-		<edge from-layer="711" from-port="0" to-layer="712" to-port="4" />
-		<edge from-layer="712" from-port="5" to-layer="717" to-port="0" />
-		<edge from-layer="713" from-port="0" to-layer="714" to-port="0" />
-		<edge from-layer="714" from-port="1" to-layer="716" to-port="0" />
-		<edge from-layer="715" from-port="0" to-layer="716" to-port="1" />
-		<edge from-layer="716" from-port="2" to-layer="717" to-port="1" />
-		<edge from-layer="717" from-port="2" to-layer="719" to-port="0" />
-		<edge from-layer="718" from-port="0" to-layer="719" to-port="1" />
-		<edge from-layer="719" from-port="2" to-layer="720" to-port="0" />
-		<edge from-layer="720" from-port="1" to-layer="722" to-port="0" />
-		<edge from-layer="721" from-port="0" to-layer="722" to-port="1" />
-		<edge from-layer="722" from-port="2" to-layer="727" to-port="0" />
-		<edge from-layer="723" from-port="0" to-layer="727" to-port="1" />
-		<edge from-layer="724" from-port="0" to-layer="727" to-port="2" />
-		<edge from-layer="725" from-port="0" to-layer="727" to-port="3" />
-		<edge from-layer="726" from-port="0" to-layer="727" to-port="4" />
-		<edge from-layer="727" from-port="5" to-layer="732" to-port="0" />
-		<edge from-layer="728" from-port="0" to-layer="729" to-port="0" />
-		<edge from-layer="729" from-port="1" to-layer="731" to-port="0" />
+		<edge from-layer="695" from-port="2" to-layer="697" to-port="0" />
+		<edge from-layer="696" from-port="0" to-layer="697" to-port="1" />
+		<edge from-layer="697" from-port="2" to-layer="713" to-port="1" />
+		<edge from-layer="698" from-port="0" to-layer="699" to-port="0" />
+		<edge from-layer="699" from-port="1" to-layer="702" to-port="0" />
+		<edge from-layer="700" from-port="0" to-layer="701" to-port="0" />
+		<edge from-layer="701" from-port="1" to-layer="702" to-port="1" />
+		<edge from-layer="702" from-port="2" to-layer="704" to-port="0" />
+		<edge from-layer="703" from-port="0" to-layer="704" to-port="1" />
+		<edge from-layer="704" from-port="2" to-layer="705" to-port="0" />
+		<edge from-layer="705" from-port="1" to-layer="706" to-port="1" />
+		<edge from-layer="706" from-port="2" to-layer="708" to-port="0" />
+		<edge from-layer="707" from-port="0" to-layer="708" to-port="1" />
+		<edge from-layer="708" from-port="2" to-layer="710" to-port="0" />
+		<edge from-layer="709" from-port="0" to-layer="710" to-port="1" />
+		<edge from-layer="710" from-port="2" to-layer="712" to-port="0" />
+		<edge from-layer="711" from-port="0" to-layer="712" to-port="1" />
+		<edge from-layer="712" from-port="2" to-layer="713" to-port="2" />
+		<edge from-layer="713" from-port="4" to-layer="715" to-port="0" />
+		<edge from-layer="714" from-port="0" to-layer="715" to-port="1" />
+		<edge from-layer="715" from-port="2" to-layer="717" to-port="0" />
+		<edge from-layer="716" from-port="0" to-layer="717" to-port="1" />
+		<edge from-layer="717" from-port="2" to-layer="726" to-port="0" />
+		<edge from-layer="718" from-port="0" to-layer="719" to-port="0" />
+		<edge from-layer="719" from-port="1" to-layer="722" to-port="0" />
+		<edge from-layer="720" from-port="0" to-layer="721" to-port="0" />
+		<edge from-layer="721" from-port="1" to-layer="722" to-port="1" />
+		<edge from-layer="722" from-port="2" to-layer="724" to-port="0" />
+		<edge from-layer="723" from-port="0" to-layer="724" to-port="1" />
+		<edge from-layer="724" from-port="2" to-layer="725" to-port="0" />
+		<edge from-layer="725" from-port="1" to-layer="726" to-port="1" />
+		<edge from-layer="726" from-port="2" to-layer="728" to-port="0" />
+		<edge from-layer="727" from-port="0" to-layer="728" to-port="1" />
+		<edge from-layer="728" from-port="2" to-layer="729" to-port="0" />
+		<edge from-layer="729" from-port="2" to-layer="731" to-port="0" />
 		<edge from-layer="730" from-port="0" to-layer="731" to-port="1" />
-		<edge from-layer="731" from-port="2" to-layer="732" to-port="1" />
-		<edge from-layer="732" from-port="2" to-layer="734" to-port="0" />
-		<edge from-layer="733" from-port="0" to-layer="734" to-port="1" />
-		<edge from-layer="734" from-port="2" to-layer="735" to-port="0" />
-		<edge from-layer="735" from-port="2" to-layer="737" to-port="0" />
-		<edge from-layer="736" from-port="0" to-layer="737" to-port="1" />
-		<edge from-layer="737" from-port="2" to-layer="739" to-port="0" />
-		<edge from-layer="738" from-port="0" to-layer="739" to-port="1" />
-		<edge from-layer="739" from-port="2" to-layer="741" to-port="0" />
-		<edge from-layer="740" from-port="0" to-layer="741" to-port="1" />
-		<edge from-layer="741" from-port="2" to-layer="811" to-port="1" />
-		<edge from-layer="741" from-port="2" to-layer="743" to-port="0" />
-		<edge from-layer="742" from-port="0" to-layer="743" to-port="1" />
-		<edge from-layer="743" from-port="2" to-layer="748" to-port="0" />
-		<edge from-layer="744" from-port="0" to-layer="748" to-port="1" />
-		<edge from-layer="745" from-port="0" to-layer="748" to-port="2" />
-		<edge from-layer="746" from-port="0" to-layer="748" to-port="3" />
-		<edge from-layer="747" from-port="0" to-layer="748" to-port="4" />
-		<edge from-layer="748" from-port="5" to-layer="753" to-port="0" />
-		<edge from-layer="748" from-port="5" to-layer="785" to-port="0" />
-		<edge from-layer="748" from-port="5" to-layer="769" to-port="0" />
-		<edge from-layer="749" from-port="0" to-layer="750" to-port="0" />
-		<edge from-layer="750" from-port="1" to-layer="752" to-port="0" />
-		<edge from-layer="751" from-port="0" to-layer="752" to-port="1" />
-		<edge from-layer="752" from-port="2" to-layer="753" to-port="1" />
-		<edge from-layer="753" from-port="2" to-layer="755" to-port="0" />
-		<edge from-layer="754" from-port="0" to-layer="755" to-port="1" />
-		<edge from-layer="755" from-port="2" to-layer="760" to-port="0" />
-		<edge from-layer="756" from-port="0" to-layer="760" to-port="1" />
-		<edge from-layer="757" from-port="0" to-layer="760" to-port="2" />
-		<edge from-layer="758" from-port="0" to-layer="760" to-port="3" />
-		<edge from-layer="759" from-port="0" to-layer="760" to-port="4" />
-		<edge from-layer="760" from-port="5" to-layer="762" to-port="0" />
-		<edge from-layer="761" from-port="0" to-layer="762" to-port="1" />
-		<edge from-layer="762" from-port="2" to-layer="764" to-port="0" />
-		<edge from-layer="763" from-port="0" to-layer="764" to-port="1" />
-		<edge from-layer="764" from-port="2" to-layer="792" to-port="0" />
-		<edge from-layer="765" from-port="0" to-layer="766" to-port="0" />
-		<edge from-layer="766" from-port="1" to-layer="768" to-port="0" />
-		<edge from-layer="767" from-port="0" to-layer="768" to-port="1" />
-		<edge from-layer="768" from-port="2" to-layer="769" to-port="1" />
-		<edge from-layer="769" from-port="2" to-layer="771" to-port="0" />
-		<edge from-layer="770" from-port="0" to-layer="771" to-port="1" />
-		<edge from-layer="771" from-port="2" to-layer="776" to-port="0" />
-		<edge from-layer="772" from-port="0" to-layer="776" to-port="1" />
-		<edge from-layer="773" from-port="0" to-layer="776" to-port="2" />
-		<edge from-layer="774" from-port="0" to-layer="776" to-port="3" />
-		<edge from-layer="775" from-port="0" to-layer="776" to-port="4" />
-		<edge from-layer="776" from-port="5" to-layer="778" to-port="0" />
+		<edge from-layer="731" from-port="2" to-layer="733" to-port="0" />
+		<edge from-layer="732" from-port="0" to-layer="733" to-port="1" />
+		<edge from-layer="733" from-port="2" to-layer="735" to-port="0" />
+		<edge from-layer="734" from-port="0" to-layer="735" to-port="1" />
+		<edge from-layer="735" from-port="2" to-layer="759" to-port="1" />
+		<edge from-layer="735" from-port="2" to-layer="744" to-port="0" />
+		<edge from-layer="736" from-port="0" to-layer="737" to-port="0" />
+		<edge from-layer="737" from-port="1" to-layer="740" to-port="0" />
+		<edge from-layer="738" from-port="0" to-layer="739" to-port="0" />
+		<edge from-layer="739" from-port="1" to-layer="740" to-port="1" />
+		<edge from-layer="740" from-port="2" to-layer="742" to-port="0" />
+		<edge from-layer="741" from-port="0" to-layer="742" to-port="1" />
+		<edge from-layer="742" from-port="2" to-layer="743" to-port="0" />
+		<edge from-layer="743" from-port="1" to-layer="744" to-port="1" />
+		<edge from-layer="744" from-port="2" to-layer="746" to-port="0" />
+		<edge from-layer="745" from-port="0" to-layer="746" to-port="1" />
+		<edge from-layer="746" from-port="2" to-layer="747" to-port="0" />
+		<edge from-layer="747" from-port="1" to-layer="756" to-port="0" />
+		<edge from-layer="748" from-port="0" to-layer="749" to-port="0" />
+		<edge from-layer="749" from-port="1" to-layer="752" to-port="0" />
+		<edge from-layer="750" from-port="0" to-layer="751" to-port="0" />
+		<edge from-layer="751" from-port="1" to-layer="752" to-port="1" />
+		<edge from-layer="752" from-port="2" to-layer="754" to-port="0" />
+		<edge from-layer="753" from-port="0" to-layer="754" to-port="1" />
+		<edge from-layer="754" from-port="2" to-layer="755" to-port="0" />
+		<edge from-layer="755" from-port="1" to-layer="756" to-port="1" />
+		<edge from-layer="756" from-port="2" to-layer="758" to-port="0" />
+		<edge from-layer="757" from-port="0" to-layer="758" to-port="1" />
+		<edge from-layer="758" from-port="2" to-layer="759" to-port="0" />
+		<edge from-layer="759" from-port="2" to-layer="761" to-port="0" />
+		<edge from-layer="760" from-port="0" to-layer="761" to-port="1" />
+		<edge from-layer="761" from-port="2" to-layer="763" to-port="0" />
+		<edge from-layer="762" from-port="0" to-layer="763" to-port="1" />
+		<edge from-layer="763" from-port="2" to-layer="765" to-port="0" />
+		<edge from-layer="764" from-port="0" to-layer="765" to-port="1" />
+		<edge from-layer="765" from-port="2" to-layer="827" to-port="1" />
+		<edge from-layer="765" from-port="2" to-layer="804" to-port="0" />
+		<edge from-layer="765" from-port="2" to-layer="789" to-port="0" />
+		<edge from-layer="765" from-port="2" to-layer="774" to-port="0" />
+		<edge from-layer="766" from-port="0" to-layer="767" to-port="0" />
+		<edge from-layer="767" from-port="1" to-layer="770" to-port="0" />
+		<edge from-layer="768" from-port="0" to-layer="769" to-port="0" />
+		<edge from-layer="769" from-port="1" to-layer="770" to-port="1" />
+		<edge from-layer="770" from-port="2" to-layer="772" to-port="0" />
+		<edge from-layer="771" from-port="0" to-layer="772" to-port="1" />
+		<edge from-layer="772" from-port="2" to-layer="773" to-port="0" />
+		<edge from-layer="773" from-port="1" to-layer="774" to-port="1" />
+		<edge from-layer="774" from-port="2" to-layer="776" to-port="0" />
+		<edge from-layer="775" from-port="0" to-layer="776" to-port="1" />
+		<edge from-layer="776" from-port="2" to-layer="778" to-port="0" />
 		<edge from-layer="777" from-port="0" to-layer="778" to-port="1" />
 		<edge from-layer="778" from-port="2" to-layer="780" to-port="0" />
 		<edge from-layer="779" from-port="0" to-layer="780" to-port="1" />
-		<edge from-layer="780" from-port="2" to-layer="792" to-port="1" />
+		<edge from-layer="780" from-port="2" to-layer="811" to-port="0" />
 		<edge from-layer="781" from-port="0" to-layer="782" to-port="0" />
-		<edge from-layer="782" from-port="1" to-layer="784" to-port="0" />
-		<edge from-layer="783" from-port="0" to-layer="784" to-port="1" />
-		<edge from-layer="784" from-port="2" to-layer="785" to-port="1" />
+		<edge from-layer="782" from-port="1" to-layer="785" to-port="0" />
+		<edge from-layer="783" from-port="0" to-layer="784" to-port="0" />
+		<edge from-layer="784" from-port="1" to-layer="785" to-port="1" />
 		<edge from-layer="785" from-port="2" to-layer="787" to-port="0" />
 		<edge from-layer="786" from-port="0" to-layer="787" to-port="1" />
-		<edge from-layer="787" from-port="2" to-layer="789" to-port="0" />
-		<edge from-layer="788" from-port="0" to-layer="789" to-port="1" />
+		<edge from-layer="787" from-port="2" to-layer="788" to-port="0" />
+		<edge from-layer="788" from-port="1" to-layer="789" to-port="1" />
 		<edge from-layer="789" from-port="2" to-layer="791" to-port="0" />
 		<edge from-layer="790" from-port="0" to-layer="791" to-port="1" />
-		<edge from-layer="791" from-port="2" to-layer="792" to-port="2" />
-		<edge from-layer="792" from-port="4" to-layer="794" to-port="0" />
-		<edge from-layer="793" from-port="0" to-layer="794" to-port="1" />
-		<edge from-layer="794" from-port="2" to-layer="796" to-port="0" />
-		<edge from-layer="795" from-port="0" to-layer="796" to-port="1" />
-		<edge from-layer="796" from-port="2" to-layer="798" to-port="0" />
-		<edge from-layer="797" from-port="0" to-layer="798" to-port="1" />
-		<edge from-layer="798" from-port="2" to-layer="803" to-port="0" />
-		<edge from-layer="799" from-port="0" to-layer="803" to-port="1" />
-		<edge from-layer="800" from-port="0" to-layer="803" to-port="2" />
-		<edge from-layer="801" from-port="0" to-layer="803" to-port="3" />
-		<edge from-layer="802" from-port="0" to-layer="803" to-port="4" />
-		<edge from-layer="803" from-port="5" to-layer="808" to-port="0" />
-		<edge from-layer="804" from-port="0" to-layer="805" to-port="0" />
-		<edge from-layer="805" from-port="1" to-layer="807" to-port="0" />
-		<edge from-layer="806" from-port="0" to-layer="807" to-port="1" />
-		<edge from-layer="807" from-port="2" to-layer="808" to-port="1" />
+		<edge from-layer="791" from-port="2" to-layer="793" to-port="0" />
+		<edge from-layer="792" from-port="0" to-layer="793" to-port="1" />
+		<edge from-layer="793" from-port="2" to-layer="795" to-port="0" />
+		<edge from-layer="794" from-port="0" to-layer="795" to-port="1" />
+		<edge from-layer="795" from-port="2" to-layer="811" to-port="1" />
+		<edge from-layer="796" from-port="0" to-layer="797" to-port="0" />
+		<edge from-layer="797" from-port="1" to-layer="800" to-port="0" />
+		<edge from-layer="798" from-port="0" to-layer="799" to-port="0" />
+		<edge from-layer="799" from-port="1" to-layer="800" to-port="1" />
+		<edge from-layer="800" from-port="2" to-layer="802" to-port="0" />
+		<edge from-layer="801" from-port="0" to-layer="802" to-port="1" />
+		<edge from-layer="802" from-port="2" to-layer="803" to-port="0" />
+		<edge from-layer="803" from-port="1" to-layer="804" to-port="1" />
+		<edge from-layer="804" from-port="2" to-layer="806" to-port="0" />
+		<edge from-layer="805" from-port="0" to-layer="806" to-port="1" />
+		<edge from-layer="806" from-port="2" to-layer="808" to-port="0" />
+		<edge from-layer="807" from-port="0" to-layer="808" to-port="1" />
 		<edge from-layer="808" from-port="2" to-layer="810" to-port="0" />
 		<edge from-layer="809" from-port="0" to-layer="810" to-port="1" />
-		<edge from-layer="810" from-port="2" to-layer="811" to-port="0" />
-		<edge from-layer="811" from-port="2" to-layer="813" to-port="0" />
+		<edge from-layer="810" from-port="2" to-layer="811" to-port="2" />
+		<edge from-layer="811" from-port="4" to-layer="813" to-port="0" />
 		<edge from-layer="812" from-port="0" to-layer="813" to-port="1" />
 		<edge from-layer="813" from-port="2" to-layer="815" to-port="0" />
 		<edge from-layer="814" from-port="0" to-layer="815" to-port="1" />
-		<edge from-layer="815" from-port="2" to-layer="817" to-port="0" />
-		<edge from-layer="816" from-port="0" to-layer="817" to-port="1" />
-		<edge from-layer="817" from-port="2" to-layer="819" to-port="0" />
-		<edge from-layer="817" from-port="2" to-layer="847" to-port="1" />
-		<edge from-layer="818" from-port="0" to-layer="819" to-port="1" />
-		<edge from-layer="819" from-port="2" to-layer="824" to-port="0" />
-		<edge from-layer="820" from-port="0" to-layer="824" to-port="1" />
-		<edge from-layer="821" from-port="0" to-layer="824" to-port="2" />
-		<edge from-layer="822" from-port="0" to-layer="824" to-port="3" />
-		<edge from-layer="823" from-port="0" to-layer="824" to-port="4" />
-		<edge from-layer="824" from-port="5" to-layer="829" to-port="0" />
-		<edge from-layer="825" from-port="0" to-layer="826" to-port="0" />
-		<edge from-layer="826" from-port="1" to-layer="828" to-port="0" />
-		<edge from-layer="827" from-port="0" to-layer="828" to-port="1" />
-		<edge from-layer="828" from-port="2" to-layer="829" to-port="1" />
+		<edge from-layer="815" from-port="2" to-layer="824" to-port="0" />
+		<edge from-layer="816" from-port="0" to-layer="817" to-port="0" />
+		<edge from-layer="817" from-port="1" to-layer="820" to-port="0" />
+		<edge from-layer="818" from-port="0" to-layer="819" to-port="0" />
+		<edge from-layer="819" from-port="1" to-layer="820" to-port="1" />
+		<edge from-layer="820" from-port="2" to-layer="822" to-port="0" />
+		<edge from-layer="821" from-port="0" to-layer="822" to-port="1" />
+		<edge from-layer="822" from-port="2" to-layer="823" to-port="0" />
+		<edge from-layer="823" from-port="1" to-layer="824" to-port="1" />
+		<edge from-layer="824" from-port="2" to-layer="826" to-port="0" />
+		<edge from-layer="825" from-port="0" to-layer="826" to-port="1" />
+		<edge from-layer="826" from-port="2" to-layer="827" to-port="0" />
+		<edge from-layer="827" from-port="2" to-layer="829" to-port="0" />
+		<edge from-layer="828" from-port="0" to-layer="829" to-port="1" />
 		<edge from-layer="829" from-port="2" to-layer="831" to-port="0" />
 		<edge from-layer="830" from-port="0" to-layer="831" to-port="1" />
-		<edge from-layer="831" from-port="2" to-layer="832" to-port="0" />
-		<edge from-layer="832" from-port="1" to-layer="834" to-port="0" />
-		<edge from-layer="833" from-port="0" to-layer="834" to-port="1" />
-		<edge from-layer="834" from-port="2" to-layer="839" to-port="0" />
-		<edge from-layer="835" from-port="0" to-layer="839" to-port="1" />
-		<edge from-layer="836" from-port="0" to-layer="839" to-port="2" />
-		<edge from-layer="837" from-port="0" to-layer="839" to-port="3" />
-		<edge from-layer="838" from-port="0" to-layer="839" to-port="4" />
-		<edge from-layer="839" from-port="5" to-layer="844" to-port="0" />
-		<edge from-layer="840" from-port="0" to-layer="841" to-port="0" />
-		<edge from-layer="841" from-port="1" to-layer="843" to-port="0" />
-		<edge from-layer="842" from-port="0" to-layer="843" to-port="1" />
-		<edge from-layer="843" from-port="2" to-layer="844" to-port="1" />
-		<edge from-layer="844" from-port="2" to-layer="846" to-port="0" />
-		<edge from-layer="845" from-port="0" to-layer="846" to-port="1" />
-		<edge from-layer="846" from-port="2" to-layer="847" to-port="0" />
-		<edge from-layer="847" from-port="2" to-layer="849" to-port="0" />
-		<edge from-layer="848" from-port="0" to-layer="849" to-port="1" />
-		<edge from-layer="849" from-port="2" to-layer="851" to-port="0" />
-		<edge from-layer="850" from-port="0" to-layer="851" to-port="1" />
-		<edge from-layer="851" from-port="2" to-layer="853" to-port="0" />
-		<edge from-layer="852" from-port="0" to-layer="853" to-port="1" />
-		<edge from-layer="853" from-port="2" to-layer="923" to-port="1" />
-		<edge from-layer="853" from-port="2" to-layer="855" to-port="0" />
-		<edge from-layer="854" from-port="0" to-layer="855" to-port="1" />
-		<edge from-layer="855" from-port="2" to-layer="860" to-port="0" />
-		<edge from-layer="856" from-port="0" to-layer="860" to-port="1" />
-		<edge from-layer="857" from-port="0" to-layer="860" to-port="2" />
-		<edge from-layer="858" from-port="0" to-layer="860" to-port="3" />
-		<edge from-layer="859" from-port="0" to-layer="860" to-port="4" />
-		<edge from-layer="860" from-port="5" to-layer="865" to-port="0" />
-		<edge from-layer="860" from-port="5" to-layer="897" to-port="0" />
-		<edge from-layer="860" from-port="5" to-layer="881" to-port="0" />
-		<edge from-layer="861" from-port="0" to-layer="862" to-port="0" />
-		<edge from-layer="862" from-port="1" to-layer="864" to-port="0" />
-		<edge from-layer="863" from-port="0" to-layer="864" to-port="1" />
-		<edge from-layer="864" from-port="2" to-layer="865" to-port="1" />
-		<edge from-layer="865" from-port="2" to-layer="867" to-port="0" />
-		<edge from-layer="866" from-port="0" to-layer="867" to-port="1" />
-		<edge from-layer="867" from-port="2" to-layer="872" to-port="0" />
-		<edge from-layer="868" from-port="0" to-layer="872" to-port="1" />
-		<edge from-layer="869" from-port="0" to-layer="872" to-port="2" />
-		<edge from-layer="870" from-port="0" to-layer="872" to-port="3" />
-		<edge from-layer="871" from-port="0" to-layer="872" to-port="4" />
-		<edge from-layer="872" from-port="5" to-layer="874" to-port="0" />
+		<edge from-layer="831" from-port="2" to-layer="833" to-port="0" />
+		<edge from-layer="832" from-port="0" to-layer="833" to-port="1" />
+		<edge from-layer="833" from-port="2" to-layer="857" to-port="1" />
+		<edge from-layer="833" from-port="2" to-layer="842" to-port="0" />
+		<edge from-layer="834" from-port="0" to-layer="835" to-port="0" />
+		<edge from-layer="835" from-port="1" to-layer="838" to-port="0" />
+		<edge from-layer="836" from-port="0" to-layer="837" to-port="0" />
+		<edge from-layer="837" from-port="1" to-layer="838" to-port="1" />
+		<edge from-layer="838" from-port="2" to-layer="840" to-port="0" />
+		<edge from-layer="839" from-port="0" to-layer="840" to-port="1" />
+		<edge from-layer="840" from-port="2" to-layer="841" to-port="0" />
+		<edge from-layer="841" from-port="1" to-layer="842" to-port="1" />
+		<edge from-layer="842" from-port="2" to-layer="844" to-port="0" />
+		<edge from-layer="843" from-port="0" to-layer="844" to-port="1" />
+		<edge from-layer="844" from-port="2" to-layer="845" to-port="0" />
+		<edge from-layer="845" from-port="1" to-layer="854" to-port="0" />
+		<edge from-layer="846" from-port="0" to-layer="847" to-port="0" />
+		<edge from-layer="847" from-port="1" to-layer="850" to-port="0" />
+		<edge from-layer="848" from-port="0" to-layer="849" to-port="0" />
+		<edge from-layer="849" from-port="1" to-layer="850" to-port="1" />
+		<edge from-layer="850" from-port="2" to-layer="852" to-port="0" />
+		<edge from-layer="851" from-port="0" to-layer="852" to-port="1" />
+		<edge from-layer="852" from-port="2" to-layer="853" to-port="0" />
+		<edge from-layer="853" from-port="1" to-layer="854" to-port="1" />
+		<edge from-layer="854" from-port="2" to-layer="856" to-port="0" />
+		<edge from-layer="855" from-port="0" to-layer="856" to-port="1" />
+		<edge from-layer="856" from-port="2" to-layer="857" to-port="0" />
+		<edge from-layer="857" from-port="2" to-layer="859" to-port="0" />
+		<edge from-layer="858" from-port="0" to-layer="859" to-port="1" />
+		<edge from-layer="859" from-port="2" to-layer="861" to-port="0" />
+		<edge from-layer="860" from-port="0" to-layer="861" to-port="1" />
+		<edge from-layer="861" from-port="2" to-layer="863" to-port="0" />
+		<edge from-layer="862" from-port="0" to-layer="863" to-port="1" />
+		<edge from-layer="863" from-port="2" to-layer="887" to-port="0" />
+		<edge from-layer="863" from-port="2" to-layer="872" to-port="0" />
+		<edge from-layer="863" from-port="2" to-layer="925" to-port="1" />
+		<edge from-layer="863" from-port="2" to-layer="902" to-port="0" />
+		<edge from-layer="864" from-port="0" to-layer="865" to-port="0" />
+		<edge from-layer="865" from-port="1" to-layer="868" to-port="0" />
+		<edge from-layer="866" from-port="0" to-layer="867" to-port="0" />
+		<edge from-layer="867" from-port="1" to-layer="868" to-port="1" />
+		<edge from-layer="868" from-port="2" to-layer="870" to-port="0" />
+		<edge from-layer="869" from-port="0" to-layer="870" to-port="1" />
+		<edge from-layer="870" from-port="2" to-layer="871" to-port="0" />
+		<edge from-layer="871" from-port="1" to-layer="872" to-port="1" />
+		<edge from-layer="872" from-port="2" to-layer="874" to-port="0" />
 		<edge from-layer="873" from-port="0" to-layer="874" to-port="1" />
 		<edge from-layer="874" from-port="2" to-layer="876" to-port="0" />
 		<edge from-layer="875" from-port="0" to-layer="876" to-port="1" />
-		<edge from-layer="876" from-port="2" to-layer="904" to-port="0" />
-		<edge from-layer="877" from-port="0" to-layer="878" to-port="0" />
-		<edge from-layer="878" from-port="1" to-layer="880" to-port="0" />
-		<edge from-layer="879" from-port="0" to-layer="880" to-port="1" />
-		<edge from-layer="880" from-port="2" to-layer="881" to-port="1" />
-		<edge from-layer="881" from-port="2" to-layer="883" to-port="0" />
-		<edge from-layer="882" from-port="0" to-layer="883" to-port="1" />
-		<edge from-layer="883" from-port="2" to-layer="888" to-port="0" />
-		<edge from-layer="884" from-port="0" to-layer="888" to-port="1" />
-		<edge from-layer="885" from-port="0" to-layer="888" to-port="2" />
-		<edge from-layer="886" from-port="0" to-layer="888" to-port="3" />
-		<edge from-layer="887" from-port="0" to-layer="888" to-port="4" />
-		<edge from-layer="888" from-port="5" to-layer="890" to-port="0" />
-		<edge from-layer="889" from-port="0" to-layer="890" to-port="1" />
-		<edge from-layer="890" from-port="2" to-layer="892" to-port="0" />
-		<edge from-layer="891" from-port="0" to-layer="892" to-port="1" />
-		<edge from-layer="892" from-port="2" to-layer="904" to-port="1" />
-		<edge from-layer="893" from-port="0" to-layer="894" to-port="0" />
-		<edge from-layer="894" from-port="1" to-layer="896" to-port="0" />
-		<edge from-layer="895" from-port="0" to-layer="896" to-port="1" />
-		<edge from-layer="896" from-port="2" to-layer="897" to-port="1" />
-		<edge from-layer="897" from-port="2" to-layer="899" to-port="0" />
-		<edge from-layer="898" from-port="0" to-layer="899" to-port="1" />
-		<edge from-layer="899" from-port="2" to-layer="901" to-port="0" />
-		<edge from-layer="900" from-port="0" to-layer="901" to-port="1" />
-		<edge from-layer="901" from-port="2" to-layer="903" to-port="0" />
-		<edge from-layer="902" from-port="0" to-layer="903" to-port="1" />
-		<edge from-layer="903" from-port="2" to-layer="904" to-port="2" />
-		<edge from-layer="904" from-port="4" to-layer="906" to-port="0" />
+		<edge from-layer="876" from-port="2" to-layer="878" to-port="0" />
+		<edge from-layer="877" from-port="0" to-layer="878" to-port="1" />
+		<edge from-layer="878" from-port="2" to-layer="909" to-port="0" />
+		<edge from-layer="879" from-port="0" to-layer="880" to-port="0" />
+		<edge from-layer="880" from-port="1" to-layer="883" to-port="0" />
+		<edge from-layer="881" from-port="0" to-layer="882" to-port="0" />
+		<edge from-layer="882" from-port="1" to-layer="883" to-port="1" />
+		<edge from-layer="883" from-port="2" to-layer="885" to-port="0" />
+		<edge from-layer="884" from-port="0" to-layer="885" to-port="1" />
+		<edge from-layer="885" from-port="2" to-layer="886" to-port="0" />
+		<edge from-layer="886" from-port="1" to-layer="887" to-port="1" />
+		<edge from-layer="887" from-port="2" to-layer="889" to-port="0" />
+		<edge from-layer="888" from-port="0" to-layer="889" to-port="1" />
+		<edge from-layer="889" from-port="2" to-layer="891" to-port="0" />
+		<edge from-layer="890" from-port="0" to-layer="891" to-port="1" />
+		<edge from-layer="891" from-port="2" to-layer="893" to-port="0" />
+		<edge from-layer="892" from-port="0" to-layer="893" to-port="1" />
+		<edge from-layer="893" from-port="2" to-layer="909" to-port="1" />
+		<edge from-layer="894" from-port="0" to-layer="895" to-port="0" />
+		<edge from-layer="895" from-port="1" to-layer="898" to-port="0" />
+		<edge from-layer="896" from-port="0" to-layer="897" to-port="0" />
+		<edge from-layer="897" from-port="1" to-layer="898" to-port="1" />
+		<edge from-layer="898" from-port="2" to-layer="900" to-port="0" />
+		<edge from-layer="899" from-port="0" to-layer="900" to-port="1" />
+		<edge from-layer="900" from-port="2" to-layer="901" to-port="0" />
+		<edge from-layer="901" from-port="1" to-layer="902" to-port="1" />
+		<edge from-layer="902" from-port="2" to-layer="904" to-port="0" />
+		<edge from-layer="903" from-port="0" to-layer="904" to-port="1" />
+		<edge from-layer="904" from-port="2" to-layer="906" to-port="0" />
 		<edge from-layer="905" from-port="0" to-layer="906" to-port="1" />
 		<edge from-layer="906" from-port="2" to-layer="908" to-port="0" />
 		<edge from-layer="907" from-port="0" to-layer="908" to-port="1" />
-		<edge from-layer="908" from-port="2" to-layer="910" to-port="0" />
-		<edge from-layer="909" from-port="0" to-layer="910" to-port="1" />
-		<edge from-layer="910" from-port="2" to-layer="915" to-port="0" />
-		<edge from-layer="911" from-port="0" to-layer="915" to-port="1" />
-		<edge from-layer="912" from-port="0" to-layer="915" to-port="2" />
-		<edge from-layer="913" from-port="0" to-layer="915" to-port="3" />
-		<edge from-layer="914" from-port="0" to-layer="915" to-port="4" />
-		<edge from-layer="915" from-port="5" to-layer="920" to-port="0" />
+		<edge from-layer="908" from-port="2" to-layer="909" to-port="2" />
+		<edge from-layer="909" from-port="4" to-layer="911" to-port="0" />
+		<edge from-layer="910" from-port="0" to-layer="911" to-port="1" />
+		<edge from-layer="911" from-port="2" to-layer="913" to-port="0" />
+		<edge from-layer="912" from-port="0" to-layer="913" to-port="1" />
+		<edge from-layer="913" from-port="2" to-layer="922" to-port="0" />
+		<edge from-layer="914" from-port="0" to-layer="915" to-port="0" />
+		<edge from-layer="915" from-port="1" to-layer="918" to-port="0" />
 		<edge from-layer="916" from-port="0" to-layer="917" to-port="0" />
-		<edge from-layer="917" from-port="1" to-layer="919" to-port="0" />
-		<edge from-layer="918" from-port="0" to-layer="919" to-port="1" />
-		<edge from-layer="919" from-port="2" to-layer="920" to-port="1" />
-		<edge from-layer="920" from-port="2" to-layer="922" to-port="0" />
-		<edge from-layer="921" from-port="0" to-layer="922" to-port="1" />
-		<edge from-layer="922" from-port="2" to-layer="923" to-port="0" />
-		<edge from-layer="923" from-port="2" to-layer="925" to-port="0" />
-		<edge from-layer="924" from-port="0" to-layer="925" to-port="1" />
+		<edge from-layer="917" from-port="1" to-layer="918" to-port="1" />
+		<edge from-layer="918" from-port="2" to-layer="920" to-port="0" />
+		<edge from-layer="919" from-port="0" to-layer="920" to-port="1" />
+		<edge from-layer="920" from-port="2" to-layer="921" to-port="0" />
+		<edge from-layer="921" from-port="1" to-layer="922" to-port="1" />
+		<edge from-layer="922" from-port="2" to-layer="924" to-port="0" />
+		<edge from-layer="923" from-port="0" to-layer="924" to-port="1" />
+		<edge from-layer="924" from-port="2" to-layer="925" to-port="0" />
 		<edge from-layer="925" from-port="2" to-layer="927" to-port="0" />
 		<edge from-layer="926" from-port="0" to-layer="927" to-port="1" />
 		<edge from-layer="927" from-port="2" to-layer="929" to-port="0" />
 		<edge from-layer="928" from-port="0" to-layer="929" to-port="1" />
-		<edge from-layer="929" from-port="2" to-layer="959" to-port="1" />
 		<edge from-layer="929" from-port="2" to-layer="931" to-port="0" />
 		<edge from-layer="930" from-port="0" to-layer="931" to-port="1" />
-		<edge from-layer="931" from-port="2" to-layer="936" to-port="0" />
-		<edge from-layer="932" from-port="0" to-layer="936" to-port="1" />
-		<edge from-layer="933" from-port="0" to-layer="936" to-port="2" />
-		<edge from-layer="934" from-port="0" to-layer="936" to-port="3" />
-		<edge from-layer="935" from-port="0" to-layer="936" to-port="4" />
-		<edge from-layer="936" from-port="5" to-layer="941" to-port="0" />
-		<edge from-layer="937" from-port="0" to-layer="938" to-port="0" />
-		<edge from-layer="938" from-port="1" to-layer="940" to-port="0" />
-		<edge from-layer="939" from-port="0" to-layer="940" to-port="1" />
-		<edge from-layer="940" from-port="2" to-layer="941" to-port="1" />
-		<edge from-layer="941" from-port="2" to-layer="943" to-port="0" />
-		<edge from-layer="942" from-port="0" to-layer="943" to-port="1" />
-		<edge from-layer="943" from-port="2" to-layer="944" to-port="0" />
-		<edge from-layer="944" from-port="1" to-layer="946" to-port="0" />
-		<edge from-layer="945" from-port="0" to-layer="946" to-port="1" />
-		<edge from-layer="946" from-port="2" to-layer="951" to-port="0" />
-		<edge from-layer="947" from-port="0" to-layer="951" to-port="1" />
-		<edge from-layer="948" from-port="0" to-layer="951" to-port="2" />
-		<edge from-layer="949" from-port="0" to-layer="951" to-port="3" />
-		<edge from-layer="950" from-port="0" to-layer="951" to-port="4" />
-		<edge from-layer="951" from-port="5" to-layer="956" to-port="0" />
-		<edge from-layer="952" from-port="0" to-layer="953" to-port="0" />
-		<edge from-layer="953" from-port="1" to-layer="955" to-port="0" />
-		<edge from-layer="954" from-port="0" to-layer="955" to-port="1" />
-		<edge from-layer="955" from-port="2" to-layer="956" to-port="1" />
-		<edge from-layer="956" from-port="2" to-layer="958" to-port="0" />
-		<edge from-layer="957" from-port="0" to-layer="958" to-port="1" />
-		<edge from-layer="958" from-port="2" to-layer="959" to-port="0" />
+		<edge from-layer="931" from-port="2" to-layer="955" to-port="1" />
+		<edge from-layer="931" from-port="2" to-layer="940" to-port="0" />
+		<edge from-layer="932" from-port="0" to-layer="933" to-port="0" />
+		<edge from-layer="933" from-port="1" to-layer="936" to-port="0" />
+		<edge from-layer="934" from-port="0" to-layer="935" to-port="0" />
+		<edge from-layer="935" from-port="1" to-layer="936" to-port="1" />
+		<edge from-layer="936" from-port="2" to-layer="938" to-port="0" />
+		<edge from-layer="937" from-port="0" to-layer="938" to-port="1" />
+		<edge from-layer="938" from-port="2" to-layer="939" to-port="0" />
+		<edge from-layer="939" from-port="1" to-layer="940" to-port="1" />
+		<edge from-layer="940" from-port="2" to-layer="942" to-port="0" />
+		<edge from-layer="941" from-port="0" to-layer="942" to-port="1" />
+		<edge from-layer="942" from-port="2" to-layer="943" to-port="0" />
+		<edge from-layer="943" from-port="1" to-layer="952" to-port="0" />
+		<edge from-layer="944" from-port="0" to-layer="945" to-port="0" />
+		<edge from-layer="945" from-port="1" to-layer="948" to-port="0" />
+		<edge from-layer="946" from-port="0" to-layer="947" to-port="0" />
+		<edge from-layer="947" from-port="1" to-layer="948" to-port="1" />
+		<edge from-layer="948" from-port="2" to-layer="950" to-port="0" />
+		<edge from-layer="949" from-port="0" to-layer="950" to-port="1" />
+		<edge from-layer="950" from-port="2" to-layer="951" to-port="0" />
+		<edge from-layer="951" from-port="1" to-layer="952" to-port="1" />
+		<edge from-layer="952" from-port="2" to-layer="954" to-port="0" />
+		<edge from-layer="953" from-port="0" to-layer="954" to-port="1" />
+		<edge from-layer="954" from-port="2" to-layer="955" to-port="0" />
+		<edge from-layer="955" from-port="2" to-layer="957" to-port="0" />
+		<edge from-layer="956" from-port="0" to-layer="957" to-port="1" />
+		<edge from-layer="957" from-port="2" to-layer="959" to-port="0" />
+		<edge from-layer="958" from-port="0" to-layer="959" to-port="1" />
 		<edge from-layer="959" from-port="2" to-layer="961" to-port="0" />
 		<edge from-layer="960" from-port="0" to-layer="961" to-port="1" />
-		<edge from-layer="961" from-port="2" to-layer="963" to-port="0" />
-		<edge from-layer="962" from-port="0" to-layer="963" to-port="1" />
-		<edge from-layer="963" from-port="2" to-layer="965" to-port="0" />
-		<edge from-layer="964" from-port="0" to-layer="965" to-port="1" />
-		<edge from-layer="965" from-port="2" to-layer="967" to-port="0" />
-		<edge from-layer="965" from-port="2" to-layer="1035" to-port="1" />
-		<edge from-layer="966" from-port="0" to-layer="967" to-port="1" />
-		<edge from-layer="967" from-port="2" to-layer="972" to-port="0" />
-		<edge from-layer="968" from-port="0" to-layer="972" to-port="1" />
-		<edge from-layer="969" from-port="0" to-layer="972" to-port="2" />
-		<edge from-layer="970" from-port="0" to-layer="972" to-port="3" />
-		<edge from-layer="971" from-port="0" to-layer="972" to-port="4" />
-		<edge from-layer="972" from-port="5" to-layer="993" to-port="0" />
-		<edge from-layer="972" from-port="5" to-layer="977" to-port="0" />
-		<edge from-layer="972" from-port="5" to-layer="1009" to-port="0" />
-		<edge from-layer="973" from-port="0" to-layer="974" to-port="0" />
-		<edge from-layer="974" from-port="1" to-layer="976" to-port="0" />
+		<edge from-layer="961" from-port="2" to-layer="1023" to-port="1" />
+		<edge from-layer="961" from-port="2" to-layer="970" to-port="0" />
+		<edge from-layer="961" from-port="2" to-layer="1000" to-port="0" />
+		<edge from-layer="961" from-port="2" to-layer="985" to-port="0" />
+		<edge from-layer="962" from-port="0" to-layer="963" to-port="0" />
+		<edge from-layer="963" from-port="1" to-layer="966" to-port="0" />
+		<edge from-layer="964" from-port="0" to-layer="965" to-port="0" />
+		<edge from-layer="965" from-port="1" to-layer="966" to-port="1" />
+		<edge from-layer="966" from-port="2" to-layer="968" to-port="0" />
+		<edge from-layer="967" from-port="0" to-layer="968" to-port="1" />
+		<edge from-layer="968" from-port="2" to-layer="969" to-port="0" />
+		<edge from-layer="969" from-port="1" to-layer="970" to-port="1" />
+		<edge from-layer="970" from-port="2" to-layer="972" to-port="0" />
+		<edge from-layer="971" from-port="0" to-layer="972" to-port="1" />
+		<edge from-layer="972" from-port="2" to-layer="974" to-port="0" />
+		<edge from-layer="973" from-port="0" to-layer="974" to-port="1" />
+		<edge from-layer="974" from-port="2" to-layer="976" to-port="0" />
 		<edge from-layer="975" from-port="0" to-layer="976" to-port="1" />
-		<edge from-layer="976" from-port="2" to-layer="977" to-port="1" />
-		<edge from-layer="977" from-port="2" to-layer="979" to-port="0" />
-		<edge from-layer="978" from-port="0" to-layer="979" to-port="1" />
-		<edge from-layer="979" from-port="2" to-layer="984" to-port="0" />
-		<edge from-layer="980" from-port="0" to-layer="984" to-port="1" />
-		<edge from-layer="981" from-port="0" to-layer="984" to-port="2" />
-		<edge from-layer="982" from-port="0" to-layer="984" to-port="3" />
-		<edge from-layer="983" from-port="0" to-layer="984" to-port="4" />
-		<edge from-layer="984" from-port="5" to-layer="986" to-port="0" />
-		<edge from-layer="985" from-port="0" to-layer="986" to-port="1" />
-		<edge from-layer="986" from-port="2" to-layer="988" to-port="0" />
-		<edge from-layer="987" from-port="0" to-layer="988" to-port="1" />
-		<edge from-layer="988" from-port="2" to-layer="1016" to-port="0" />
-		<edge from-layer="989" from-port="0" to-layer="990" to-port="0" />
-		<edge from-layer="990" from-port="1" to-layer="992" to-port="0" />
-		<edge from-layer="991" from-port="0" to-layer="992" to-port="1" />
-		<edge from-layer="992" from-port="2" to-layer="993" to-port="1" />
-		<edge from-layer="993" from-port="2" to-layer="995" to-port="0" />
-		<edge from-layer="994" from-port="0" to-layer="995" to-port="1" />
-		<edge from-layer="995" from-port="2" to-layer="1000" to-port="0" />
-		<edge from-layer="996" from-port="0" to-layer="1000" to-port="1" />
-		<edge from-layer="997" from-port="0" to-layer="1000" to-port="2" />
-		<edge from-layer="998" from-port="0" to-layer="1000" to-port="3" />
-		<edge from-layer="999" from-port="0" to-layer="1000" to-port="4" />
-		<edge from-layer="1000" from-port="5" to-layer="1002" to-port="0" />
+		<edge from-layer="976" from-port="2" to-layer="1007" to-port="0" />
+		<edge from-layer="977" from-port="0" to-layer="978" to-port="0" />
+		<edge from-layer="978" from-port="1" to-layer="981" to-port="0" />
+		<edge from-layer="979" from-port="0" to-layer="980" to-port="0" />
+		<edge from-layer="980" from-port="1" to-layer="981" to-port="1" />
+		<edge from-layer="981" from-port="2" to-layer="983" to-port="0" />
+		<edge from-layer="982" from-port="0" to-layer="983" to-port="1" />
+		<edge from-layer="983" from-port="2" to-layer="984" to-port="0" />
+		<edge from-layer="984" from-port="1" to-layer="985" to-port="1" />
+		<edge from-layer="985" from-port="2" to-layer="987" to-port="0" />
+		<edge from-layer="986" from-port="0" to-layer="987" to-port="1" />
+		<edge from-layer="987" from-port="2" to-layer="989" to-port="0" />
+		<edge from-layer="988" from-port="0" to-layer="989" to-port="1" />
+		<edge from-layer="989" from-port="2" to-layer="991" to-port="0" />
+		<edge from-layer="990" from-port="0" to-layer="991" to-port="1" />
+		<edge from-layer="991" from-port="2" to-layer="1007" to-port="1" />
+		<edge from-layer="992" from-port="0" to-layer="993" to-port="0" />
+		<edge from-layer="993" from-port="1" to-layer="996" to-port="0" />
+		<edge from-layer="994" from-port="0" to-layer="995" to-port="0" />
+		<edge from-layer="995" from-port="1" to-layer="996" to-port="1" />
+		<edge from-layer="996" from-port="2" to-layer="998" to-port="0" />
+		<edge from-layer="997" from-port="0" to-layer="998" to-port="1" />
+		<edge from-layer="998" from-port="2" to-layer="999" to-port="0" />
+		<edge from-layer="999" from-port="1" to-layer="1000" to-port="1" />
+		<edge from-layer="1000" from-port="2" to-layer="1002" to-port="0" />
 		<edge from-layer="1001" from-port="0" to-layer="1002" to-port="1" />
 		<edge from-layer="1002" from-port="2" to-layer="1004" to-port="0" />
 		<edge from-layer="1003" from-port="0" to-layer="1004" to-port="1" />
-		<edge from-layer="1004" from-port="2" to-layer="1016" to-port="1" />
-		<edge from-layer="1005" from-port="0" to-layer="1006" to-port="0" />
-		<edge from-layer="1006" from-port="1" to-layer="1008" to-port="0" />
-		<edge from-layer="1007" from-port="0" to-layer="1008" to-port="1" />
-		<edge from-layer="1008" from-port="2" to-layer="1009" to-port="1" />
+		<edge from-layer="1004" from-port="2" to-layer="1006" to-port="0" />
+		<edge from-layer="1005" from-port="0" to-layer="1006" to-port="1" />
+		<edge from-layer="1006" from-port="2" to-layer="1007" to-port="2" />
+		<edge from-layer="1007" from-port="4" to-layer="1009" to-port="0" />
+		<edge from-layer="1008" from-port="0" to-layer="1009" to-port="1" />
 		<edge from-layer="1009" from-port="2" to-layer="1011" to-port="0" />
 		<edge from-layer="1010" from-port="0" to-layer="1011" to-port="1" />
-		<edge from-layer="1011" from-port="2" to-layer="1013" to-port="0" />
-		<edge from-layer="1012" from-port="0" to-layer="1013" to-port="1" />
-		<edge from-layer="1013" from-port="2" to-layer="1015" to-port="0" />
-		<edge from-layer="1014" from-port="0" to-layer="1015" to-port="1" />
-		<edge from-layer="1015" from-port="2" to-layer="1016" to-port="2" />
-		<edge from-layer="1016" from-port="4" to-layer="1018" to-port="0" />
+		<edge from-layer="1011" from-port="2" to-layer="1020" to-port="0" />
+		<edge from-layer="1012" from-port="0" to-layer="1013" to-port="0" />
+		<edge from-layer="1013" from-port="1" to-layer="1016" to-port="0" />
+		<edge from-layer="1014" from-port="0" to-layer="1015" to-port="0" />
+		<edge from-layer="1015" from-port="1" to-layer="1016" to-port="1" />
+		<edge from-layer="1016" from-port="2" to-layer="1018" to-port="0" />
 		<edge from-layer="1017" from-port="0" to-layer="1018" to-port="1" />
-		<edge from-layer="1018" from-port="2" to-layer="1020" to-port="0" />
-		<edge from-layer="1019" from-port="0" to-layer="1020" to-port="1" />
+		<edge from-layer="1018" from-port="2" to-layer="1019" to-port="0" />
+		<edge from-layer="1019" from-port="1" to-layer="1020" to-port="1" />
 		<edge from-layer="1020" from-port="2" to-layer="1022" to-port="0" />
 		<edge from-layer="1021" from-port="0" to-layer="1022" to-port="1" />
-		<edge from-layer="1022" from-port="2" to-layer="1027" to-port="0" />
-		<edge from-layer="1023" from-port="0" to-layer="1027" to-port="1" />
-		<edge from-layer="1024" from-port="0" to-layer="1027" to-port="2" />
-		<edge from-layer="1025" from-port="0" to-layer="1027" to-port="3" />
-		<edge from-layer="1026" from-port="0" to-layer="1027" to-port="4" />
-		<edge from-layer="1027" from-port="5" to-layer="1032" to-port="0" />
-		<edge from-layer="1028" from-port="0" to-layer="1029" to-port="0" />
-		<edge from-layer="1029" from-port="1" to-layer="1031" to-port="0" />
-		<edge from-layer="1030" from-port="0" to-layer="1031" to-port="1" />
-		<edge from-layer="1031" from-port="2" to-layer="1032" to-port="1" />
-		<edge from-layer="1032" from-port="2" to-layer="1034" to-port="0" />
-		<edge from-layer="1033" from-port="0" to-layer="1034" to-port="1" />
-		<edge from-layer="1034" from-port="2" to-layer="1035" to-port="0" />
-		<edge from-layer="1035" from-port="2" to-layer="1037" to-port="0" />
-		<edge from-layer="1036" from-port="0" to-layer="1037" to-port="1" />
-		<edge from-layer="1037" from-port="2" to-layer="1039" to-port="0" />
-		<edge from-layer="1038" from-port="0" to-layer="1039" to-port="1" />
-		<edge from-layer="1039" from-port="2" to-layer="1041" to-port="0" />
-		<edge from-layer="1040" from-port="0" to-layer="1041" to-port="1" />
-		<edge from-layer="1041" from-port="2" to-layer="1043" to-port="0" />
-		<edge from-layer="1041" from-port="2" to-layer="1071" to-port="1" />
-		<edge from-layer="1042" from-port="0" to-layer="1043" to-port="1" />
-		<edge from-layer="1043" from-port="2" to-layer="1048" to-port="0" />
-		<edge from-layer="1044" from-port="0" to-layer="1048" to-port="1" />
-		<edge from-layer="1045" from-port="0" to-layer="1048" to-port="2" />
-		<edge from-layer="1046" from-port="0" to-layer="1048" to-port="3" />
-		<edge from-layer="1047" from-port="0" to-layer="1048" to-port="4" />
-		<edge from-layer="1048" from-port="5" to-layer="1053" to-port="0" />
-		<edge from-layer="1049" from-port="0" to-layer="1050" to-port="0" />
-		<edge from-layer="1050" from-port="1" to-layer="1052" to-port="0" />
+		<edge from-layer="1022" from-port="2" to-layer="1023" to-port="0" />
+		<edge from-layer="1023" from-port="2" to-layer="1025" to-port="0" />
+		<edge from-layer="1024" from-port="0" to-layer="1025" to-port="1" />
+		<edge from-layer="1025" from-port="2" to-layer="1027" to-port="0" />
+		<edge from-layer="1026" from-port="0" to-layer="1027" to-port="1" />
+		<edge from-layer="1027" from-port="2" to-layer="1029" to-port="0" />
+		<edge from-layer="1028" from-port="0" to-layer="1029" to-port="1" />
+		<edge from-layer="1029" from-port="2" to-layer="1053" to-port="1" />
+		<edge from-layer="1029" from-port="2" to-layer="1038" to-port="0" />
+		<edge from-layer="1030" from-port="0" to-layer="1031" to-port="0" />
+		<edge from-layer="1031" from-port="1" to-layer="1034" to-port="0" />
+		<edge from-layer="1032" from-port="0" to-layer="1033" to-port="0" />
+		<edge from-layer="1033" from-port="1" to-layer="1034" to-port="1" />
+		<edge from-layer="1034" from-port="2" to-layer="1036" to-port="0" />
+		<edge from-layer="1035" from-port="0" to-layer="1036" to-port="1" />
+		<edge from-layer="1036" from-port="2" to-layer="1037" to-port="0" />
+		<edge from-layer="1037" from-port="1" to-layer="1038" to-port="1" />
+		<edge from-layer="1038" from-port="2" to-layer="1040" to-port="0" />
+		<edge from-layer="1039" from-port="0" to-layer="1040" to-port="1" />
+		<edge from-layer="1040" from-port="2" to-layer="1041" to-port="0" />
+		<edge from-layer="1041" from-port="1" to-layer="1050" to-port="0" />
+		<edge from-layer="1042" from-port="0" to-layer="1043" to-port="0" />
+		<edge from-layer="1043" from-port="1" to-layer="1046" to-port="0" />
+		<edge from-layer="1044" from-port="0" to-layer="1045" to-port="0" />
+		<edge from-layer="1045" from-port="1" to-layer="1046" to-port="1" />
+		<edge from-layer="1046" from-port="2" to-layer="1048" to-port="0" />
+		<edge from-layer="1047" from-port="0" to-layer="1048" to-port="1" />
+		<edge from-layer="1048" from-port="2" to-layer="1049" to-port="0" />
+		<edge from-layer="1049" from-port="1" to-layer="1050" to-port="1" />
+		<edge from-layer="1050" from-port="2" to-layer="1052" to-port="0" />
 		<edge from-layer="1051" from-port="0" to-layer="1052" to-port="1" />
-		<edge from-layer="1052" from-port="2" to-layer="1053" to-port="1" />
+		<edge from-layer="1052" from-port="2" to-layer="1053" to-port="0" />
 		<edge from-layer="1053" from-port="2" to-layer="1055" to-port="0" />
 		<edge from-layer="1054" from-port="0" to-layer="1055" to-port="1" />
-		<edge from-layer="1055" from-port="2" to-layer="1056" to-port="0" />
-		<edge from-layer="1056" from-port="1" to-layer="1058" to-port="0" />
-		<edge from-layer="1057" from-port="0" to-layer="1058" to-port="1" />
-		<edge from-layer="1058" from-port="2" to-layer="1063" to-port="0" />
-		<edge from-layer="1059" from-port="0" to-layer="1063" to-port="1" />
-		<edge from-layer="1060" from-port="0" to-layer="1063" to-port="2" />
-		<edge from-layer="1061" from-port="0" to-layer="1063" to-port="3" />
-		<edge from-layer="1062" from-port="0" to-layer="1063" to-port="4" />
-		<edge from-layer="1063" from-port="5" to-layer="1068" to-port="0" />
-		<edge from-layer="1064" from-port="0" to-layer="1065" to-port="0" />
-		<edge from-layer="1065" from-port="1" to-layer="1067" to-port="0" />
-		<edge from-layer="1066" from-port="0" to-layer="1067" to-port="1" />
-		<edge from-layer="1067" from-port="2" to-layer="1068" to-port="1" />
+		<edge from-layer="1055" from-port="2" to-layer="1057" to-port="0" />
+		<edge from-layer="1056" from-port="0" to-layer="1057" to-port="1" />
+		<edge from-layer="1057" from-port="2" to-layer="1059" to-port="0" />
+		<edge from-layer="1058" from-port="0" to-layer="1059" to-port="1" />
+		<edge from-layer="1059" from-port="2" to-layer="1083" to-port="0" />
+		<edge from-layer="1059" from-port="2" to-layer="1121" to-port="1" />
+		<edge from-layer="1059" from-port="2" to-layer="1098" to-port="0" />
+		<edge from-layer="1059" from-port="2" to-layer="1068" to-port="0" />
+		<edge from-layer="1060" from-port="0" to-layer="1061" to-port="0" />
+		<edge from-layer="1061" from-port="1" to-layer="1064" to-port="0" />
+		<edge from-layer="1062" from-port="0" to-layer="1063" to-port="0" />
+		<edge from-layer="1063" from-port="1" to-layer="1064" to-port="1" />
+		<edge from-layer="1064" from-port="2" to-layer="1066" to-port="0" />
+		<edge from-layer="1065" from-port="0" to-layer="1066" to-port="1" />
+		<edge from-layer="1066" from-port="2" to-layer="1067" to-port="0" />
+		<edge from-layer="1067" from-port="1" to-layer="1068" to-port="1" />
 		<edge from-layer="1068" from-port="2" to-layer="1070" to-port="0" />
 		<edge from-layer="1069" from-port="0" to-layer="1070" to-port="1" />
-		<edge from-layer="1070" from-port="2" to-layer="1071" to-port="0" />
-		<edge from-layer="1071" from-port="2" to-layer="1073" to-port="0" />
-		<edge from-layer="1072" from-port="0" to-layer="1073" to-port="1" />
-		<edge from-layer="1073" from-port="2" to-layer="1075" to-port="0" />
-		<edge from-layer="1074" from-port="0" to-layer="1075" to-port="1" />
-		<edge from-layer="1075" from-port="2" to-layer="1077" to-port="0" />
-		<edge from-layer="1076" from-port="0" to-layer="1077" to-port="1" />
-		<edge from-layer="1077" from-port="2" to-layer="1079" to-port="0" />
-		<edge from-layer="1077" from-port="2" to-layer="1147" to-port="1" />
-		<edge from-layer="1078" from-port="0" to-layer="1079" to-port="1" />
-		<edge from-layer="1079" from-port="2" to-layer="1084" to-port="0" />
-		<edge from-layer="1080" from-port="0" to-layer="1084" to-port="1" />
-		<edge from-layer="1081" from-port="0" to-layer="1084" to-port="2" />
-		<edge from-layer="1082" from-port="0" to-layer="1084" to-port="3" />
-		<edge from-layer="1083" from-port="0" to-layer="1084" to-port="4" />
-		<edge from-layer="1084" from-port="5" to-layer="1089" to-port="0" />
-		<edge from-layer="1084" from-port="5" to-layer="1121" to-port="0" />
-		<edge from-layer="1084" from-port="5" to-layer="1105" to-port="0" />
-		<edge from-layer="1085" from-port="0" to-layer="1086" to-port="0" />
-		<edge from-layer="1086" from-port="1" to-layer="1088" to-port="0" />
-		<edge from-layer="1087" from-port="0" to-layer="1088" to-port="1" />
-		<edge from-layer="1088" from-port="2" to-layer="1089" to-port="1" />
-		<edge from-layer="1089" from-port="2" to-layer="1091" to-port="0" />
-		<edge from-layer="1090" from-port="0" to-layer="1091" to-port="1" />
-		<edge from-layer="1091" from-port="2" to-layer="1096" to-port="0" />
-		<edge from-layer="1092" from-port="0" to-layer="1096" to-port="1" />
-		<edge from-layer="1093" from-port="0" to-layer="1096" to-port="2" />
-		<edge from-layer="1094" from-port="0" to-layer="1096" to-port="3" />
-		<edge from-layer="1095" from-port="0" to-layer="1096" to-port="4" />
-		<edge from-layer="1096" from-port="5" to-layer="1098" to-port="0" />
-		<edge from-layer="1097" from-port="0" to-layer="1098" to-port="1" />
+		<edge from-layer="1070" from-port="2" to-layer="1072" to-port="0" />
+		<edge from-layer="1071" from-port="0" to-layer="1072" to-port="1" />
+		<edge from-layer="1072" from-port="2" to-layer="1074" to-port="0" />
+		<edge from-layer="1073" from-port="0" to-layer="1074" to-port="1" />
+		<edge from-layer="1074" from-port="2" to-layer="1105" to-port="0" />
+		<edge from-layer="1075" from-port="0" to-layer="1076" to-port="0" />
+		<edge from-layer="1076" from-port="1" to-layer="1079" to-port="0" />
+		<edge from-layer="1077" from-port="0" to-layer="1078" to-port="0" />
+		<edge from-layer="1078" from-port="1" to-layer="1079" to-port="1" />
+		<edge from-layer="1079" from-port="2" to-layer="1081" to-port="0" />
+		<edge from-layer="1080" from-port="0" to-layer="1081" to-port="1" />
+		<edge from-layer="1081" from-port="2" to-layer="1082" to-port="0" />
+		<edge from-layer="1082" from-port="1" to-layer="1083" to-port="1" />
+		<edge from-layer="1083" from-port="2" to-layer="1085" to-port="0" />
+		<edge from-layer="1084" from-port="0" to-layer="1085" to-port="1" />
+		<edge from-layer="1085" from-port="2" to-layer="1087" to-port="0" />
+		<edge from-layer="1086" from-port="0" to-layer="1087" to-port="1" />
+		<edge from-layer="1087" from-port="2" to-layer="1089" to-port="0" />
+		<edge from-layer="1088" from-port="0" to-layer="1089" to-port="1" />
+		<edge from-layer="1089" from-port="2" to-layer="1105" to-port="1" />
+		<edge from-layer="1090" from-port="0" to-layer="1091" to-port="0" />
+		<edge from-layer="1091" from-port="1" to-layer="1094" to-port="0" />
+		<edge from-layer="1092" from-port="0" to-layer="1093" to-port="0" />
+		<edge from-layer="1093" from-port="1" to-layer="1094" to-port="1" />
+		<edge from-layer="1094" from-port="2" to-layer="1096" to-port="0" />
+		<edge from-layer="1095" from-port="0" to-layer="1096" to-port="1" />
+		<edge from-layer="1096" from-port="2" to-layer="1097" to-port="0" />
+		<edge from-layer="1097" from-port="1" to-layer="1098" to-port="1" />
 		<edge from-layer="1098" from-port="2" to-layer="1100" to-port="0" />
 		<edge from-layer="1099" from-port="0" to-layer="1100" to-port="1" />
-		<edge from-layer="1100" from-port="2" to-layer="1128" to-port="0" />
-		<edge from-layer="1101" from-port="0" to-layer="1102" to-port="0" />
-		<edge from-layer="1102" from-port="1" to-layer="1104" to-port="0" />
+		<edge from-layer="1100" from-port="2" to-layer="1102" to-port="0" />
+		<edge from-layer="1101" from-port="0" to-layer="1102" to-port="1" />
+		<edge from-layer="1102" from-port="2" to-layer="1104" to-port="0" />
 		<edge from-layer="1103" from-port="0" to-layer="1104" to-port="1" />
-		<edge from-layer="1104" from-port="2" to-layer="1105" to-port="1" />
-		<edge from-layer="1105" from-port="2" to-layer="1107" to-port="0" />
+		<edge from-layer="1104" from-port="2" to-layer="1105" to-port="2" />
+		<edge from-layer="1105" from-port="4" to-layer="1107" to-port="0" />
 		<edge from-layer="1106" from-port="0" to-layer="1107" to-port="1" />
-		<edge from-layer="1107" from-port="2" to-layer="1112" to-port="0" />
-		<edge from-layer="1108" from-port="0" to-layer="1112" to-port="1" />
-		<edge from-layer="1109" from-port="0" to-layer="1112" to-port="2" />
-		<edge from-layer="1110" from-port="0" to-layer="1112" to-port="3" />
-		<edge from-layer="1111" from-port="0" to-layer="1112" to-port="4" />
-		<edge from-layer="1112" from-port="5" to-layer="1114" to-port="0" />
-		<edge from-layer="1113" from-port="0" to-layer="1114" to-port="1" />
+		<edge from-layer="1107" from-port="2" to-layer="1109" to-port="0" />
+		<edge from-layer="1108" from-port="0" to-layer="1109" to-port="1" />
+		<edge from-layer="1109" from-port="2" to-layer="1118" to-port="0" />
+		<edge from-layer="1110" from-port="0" to-layer="1111" to-port="0" />
+		<edge from-layer="1111" from-port="1" to-layer="1114" to-port="0" />
+		<edge from-layer="1112" from-port="0" to-layer="1113" to-port="0" />
+		<edge from-layer="1113" from-port="1" to-layer="1114" to-port="1" />
 		<edge from-layer="1114" from-port="2" to-layer="1116" to-port="0" />
 		<edge from-layer="1115" from-port="0" to-layer="1116" to-port="1" />
-		<edge from-layer="1116" from-port="2" to-layer="1128" to-port="1" />
-		<edge from-layer="1117" from-port="0" to-layer="1118" to-port="0" />
-		<edge from-layer="1118" from-port="1" to-layer="1120" to-port="0" />
+		<edge from-layer="1116" from-port="2" to-layer="1117" to-port="0" />
+		<edge from-layer="1117" from-port="1" to-layer="1118" to-port="1" />
+		<edge from-layer="1118" from-port="2" to-layer="1120" to-port="0" />
 		<edge from-layer="1119" from-port="0" to-layer="1120" to-port="1" />
-		<edge from-layer="1120" from-port="2" to-layer="1121" to-port="1" />
+		<edge from-layer="1120" from-port="2" to-layer="1121" to-port="0" />
 		<edge from-layer="1121" from-port="2" to-layer="1123" to-port="0" />
 		<edge from-layer="1122" from-port="0" to-layer="1123" to-port="1" />
 		<edge from-layer="1123" from-port="2" to-layer="1125" to-port="0" />
 		<edge from-layer="1124" from-port="0" to-layer="1125" to-port="1" />
 		<edge from-layer="1125" from-port="2" to-layer="1127" to-port="0" />
 		<edge from-layer="1126" from-port="0" to-layer="1127" to-port="1" />
-		<edge from-layer="1127" from-port="2" to-layer="1128" to-port="2" />
-		<edge from-layer="1128" from-port="4" to-layer="1130" to-port="0" />
-		<edge from-layer="1129" from-port="0" to-layer="1130" to-port="1" />
-		<edge from-layer="1130" from-port="2" to-layer="1132" to-port="0" />
-		<edge from-layer="1131" from-port="0" to-layer="1132" to-port="1" />
+		<edge from-layer="1127" from-port="2" to-layer="1151" to-port="1" />
+		<edge from-layer="1127" from-port="2" to-layer="1136" to-port="0" />
+		<edge from-layer="1128" from-port="0" to-layer="1129" to-port="0" />
+		<edge from-layer="1129" from-port="1" to-layer="1132" to-port="0" />
+		<edge from-layer="1130" from-port="0" to-layer="1131" to-port="0" />
+		<edge from-layer="1131" from-port="1" to-layer="1132" to-port="1" />
 		<edge from-layer="1132" from-port="2" to-layer="1134" to-port="0" />
 		<edge from-layer="1133" from-port="0" to-layer="1134" to-port="1" />
-		<edge from-layer="1134" from-port="2" to-layer="1139" to-port="0" />
-		<edge from-layer="1135" from-port="0" to-layer="1139" to-port="1" />
-		<edge from-layer="1136" from-port="0" to-layer="1139" to-port="2" />
-		<edge from-layer="1137" from-port="0" to-layer="1139" to-port="3" />
-		<edge from-layer="1138" from-port="0" to-layer="1139" to-port="4" />
-		<edge from-layer="1139" from-port="5" to-layer="1144" to-port="0" />
+		<edge from-layer="1134" from-port="2" to-layer="1135" to-port="0" />
+		<edge from-layer="1135" from-port="1" to-layer="1136" to-port="1" />
+		<edge from-layer="1136" from-port="2" to-layer="1138" to-port="0" />
+		<edge from-layer="1137" from-port="0" to-layer="1138" to-port="1" />
+		<edge from-layer="1138" from-port="2" to-layer="1139" to-port="0" />
+		<edge from-layer="1139" from-port="1" to-layer="1148" to-port="0" />
 		<edge from-layer="1140" from-port="0" to-layer="1141" to-port="0" />
-		<edge from-layer="1141" from-port="1" to-layer="1143" to-port="0" />
-		<edge from-layer="1142" from-port="0" to-layer="1143" to-port="1" />
-		<edge from-layer="1143" from-port="2" to-layer="1144" to-port="1" />
+		<edge from-layer="1141" from-port="1" to-layer="1144" to-port="0" />
+		<edge from-layer="1142" from-port="0" to-layer="1143" to-port="0" />
+		<edge from-layer="1143" from-port="1" to-layer="1144" to-port="1" />
 		<edge from-layer="1144" from-port="2" to-layer="1146" to-port="0" />
 		<edge from-layer="1145" from-port="0" to-layer="1146" to-port="1" />
 		<edge from-layer="1146" from-port="2" to-layer="1147" to-port="0" />
-		<edge from-layer="1147" from-port="2" to-layer="1149" to-port="0" />
-		<edge from-layer="1148" from-port="0" to-layer="1149" to-port="1" />
-		<edge from-layer="1149" from-port="2" to-layer="1151" to-port="0" />
-		<edge from-layer="1150" from-port="0" to-layer="1151" to-port="1" />
+		<edge from-layer="1147" from-port="1" to-layer="1148" to-port="1" />
+		<edge from-layer="1148" from-port="2" to-layer="1150" to-port="0" />
+		<edge from-layer="1149" from-port="0" to-layer="1150" to-port="1" />
+		<edge from-layer="1150" from-port="2" to-layer="1151" to-port="0" />
 		<edge from-layer="1151" from-port="2" to-layer="1153" to-port="0" />
 		<edge from-layer="1152" from-port="0" to-layer="1153" to-port="1" />
-		<edge from-layer="1153" from-port="2" to-layer="1183" to-port="1" />
 		<edge from-layer="1153" from-port="2" to-layer="1155" to-port="0" />
 		<edge from-layer="1154" from-port="0" to-layer="1155" to-port="1" />
-		<edge from-layer="1155" from-port="2" to-layer="1160" to-port="0" />
-		<edge from-layer="1156" from-port="0" to-layer="1160" to-port="1" />
-		<edge from-layer="1157" from-port="0" to-layer="1160" to-port="2" />
-		<edge from-layer="1158" from-port="0" to-layer="1160" to-port="3" />
-		<edge from-layer="1159" from-port="0" to-layer="1160" to-port="4" />
-		<edge from-layer="1160" from-port="5" to-layer="1165" to-port="0" />
-		<edge from-layer="1161" from-port="0" to-layer="1162" to-port="0" />
-		<edge from-layer="1162" from-port="1" to-layer="1164" to-port="0" />
+		<edge from-layer="1155" from-port="2" to-layer="1157" to-port="0" />
+		<edge from-layer="1156" from-port="0" to-layer="1157" to-port="1" />
+		<edge from-layer="1157" from-port="2" to-layer="1219" to-port="1" />
+		<edge from-layer="1157" from-port="2" to-layer="1196" to-port="0" />
+		<edge from-layer="1157" from-port="2" to-layer="1181" to-port="0" />
+		<edge from-layer="1157" from-port="2" to-layer="1166" to-port="0" />
+		<edge from-layer="1158" from-port="0" to-layer="1159" to-port="0" />
+		<edge from-layer="1159" from-port="1" to-layer="1162" to-port="0" />
+		<edge from-layer="1160" from-port="0" to-layer="1161" to-port="0" />
+		<edge from-layer="1161" from-port="1" to-layer="1162" to-port="1" />
+		<edge from-layer="1162" from-port="2" to-layer="1164" to-port="0" />
 		<edge from-layer="1163" from-port="0" to-layer="1164" to-port="1" />
-		<edge from-layer="1164" from-port="2" to-layer="1165" to-port="1" />
-		<edge from-layer="1165" from-port="2" to-layer="1167" to-port="0" />
-		<edge from-layer="1166" from-port="0" to-layer="1167" to-port="1" />
-		<edge from-layer="1167" from-port="2" to-layer="1168" to-port="0" />
-		<edge from-layer="1168" from-port="1" to-layer="1170" to-port="0" />
+		<edge from-layer="1164" from-port="2" to-layer="1165" to-port="0" />
+		<edge from-layer="1165" from-port="1" to-layer="1166" to-port="1" />
+		<edge from-layer="1166" from-port="2" to-layer="1168" to-port="0" />
+		<edge from-layer="1167" from-port="0" to-layer="1168" to-port="1" />
+		<edge from-layer="1168" from-port="2" to-layer="1170" to-port="0" />
 		<edge from-layer="1169" from-port="0" to-layer="1170" to-port="1" />
-		<edge from-layer="1170" from-port="2" to-layer="1175" to-port="0" />
-		<edge from-layer="1171" from-port="0" to-layer="1175" to-port="1" />
-		<edge from-layer="1172" from-port="0" to-layer="1175" to-port="2" />
-		<edge from-layer="1173" from-port="0" to-layer="1175" to-port="3" />
-		<edge from-layer="1174" from-port="0" to-layer="1175" to-port="4" />
-		<edge from-layer="1175" from-port="5" to-layer="1180" to-port="0" />
-		<edge from-layer="1176" from-port="0" to-layer="1177" to-port="0" />
-		<edge from-layer="1177" from-port="1" to-layer="1179" to-port="0" />
+		<edge from-layer="1170" from-port="2" to-layer="1172" to-port="0" />
+		<edge from-layer="1171" from-port="0" to-layer="1172" to-port="1" />
+		<edge from-layer="1172" from-port="2" to-layer="1203" to-port="0" />
+		<edge from-layer="1173" from-port="0" to-layer="1174" to-port="0" />
+		<edge from-layer="1174" from-port="1" to-layer="1177" to-port="0" />
+		<edge from-layer="1175" from-port="0" to-layer="1176" to-port="0" />
+		<edge from-layer="1176" from-port="1" to-layer="1177" to-port="1" />
+		<edge from-layer="1177" from-port="2" to-layer="1179" to-port="0" />
 		<edge from-layer="1178" from-port="0" to-layer="1179" to-port="1" />
-		<edge from-layer="1179" from-port="2" to-layer="1180" to-port="1" />
-		<edge from-layer="1180" from-port="2" to-layer="1182" to-port="0" />
-		<edge from-layer="1181" from-port="0" to-layer="1182" to-port="1" />
-		<edge from-layer="1182" from-port="2" to-layer="1183" to-port="0" />
+		<edge from-layer="1179" from-port="2" to-layer="1180" to-port="0" />
+		<edge from-layer="1180" from-port="1" to-layer="1181" to-port="1" />
+		<edge from-layer="1181" from-port="2" to-layer="1183" to-port="0" />
+		<edge from-layer="1182" from-port="0" to-layer="1183" to-port="1" />
 		<edge from-layer="1183" from-port="2" to-layer="1185" to-port="0" />
 		<edge from-layer="1184" from-port="0" to-layer="1185" to-port="1" />
 		<edge from-layer="1185" from-port="2" to-layer="1187" to-port="0" />
 		<edge from-layer="1186" from-port="0" to-layer="1187" to-port="1" />
-		<edge from-layer="1187" from-port="2" to-layer="1189" to-port="0" />
-		<edge from-layer="1188" from-port="0" to-layer="1189" to-port="1" />
-		<edge from-layer="1189" from-port="2" to-layer="1191" to-port="0" />
-		<edge from-layer="1189" from-port="2" to-layer="1259" to-port="1" />
-		<edge from-layer="1190" from-port="0" to-layer="1191" to-port="1" />
-		<edge from-layer="1191" from-port="2" to-layer="1196" to-port="0" />
-		<edge from-layer="1192" from-port="0" to-layer="1196" to-port="1" />
-		<edge from-layer="1193" from-port="0" to-layer="1196" to-port="2" />
-		<edge from-layer="1194" from-port="0" to-layer="1196" to-port="3" />
-		<edge from-layer="1195" from-port="0" to-layer="1196" to-port="4" />
-		<edge from-layer="1196" from-port="5" to-layer="1233" to-port="0" />
-		<edge from-layer="1196" from-port="5" to-layer="1217" to-port="0" />
-		<edge from-layer="1196" from-port="5" to-layer="1201" to-port="0" />
-		<edge from-layer="1197" from-port="0" to-layer="1198" to-port="0" />
-		<edge from-layer="1198" from-port="1" to-layer="1200" to-port="0" />
+		<edge from-layer="1187" from-port="2" to-layer="1203" to-port="1" />
+		<edge from-layer="1188" from-port="0" to-layer="1189" to-port="0" />
+		<edge from-layer="1189" from-port="1" to-layer="1192" to-port="0" />
+		<edge from-layer="1190" from-port="0" to-layer="1191" to-port="0" />
+		<edge from-layer="1191" from-port="1" to-layer="1192" to-port="1" />
+		<edge from-layer="1192" from-port="2" to-layer="1194" to-port="0" />
+		<edge from-layer="1193" from-port="0" to-layer="1194" to-port="1" />
+		<edge from-layer="1194" from-port="2" to-layer="1195" to-port="0" />
+		<edge from-layer="1195" from-port="1" to-layer="1196" to-port="1" />
+		<edge from-layer="1196" from-port="2" to-layer="1198" to-port="0" />
+		<edge from-layer="1197" from-port="0" to-layer="1198" to-port="1" />
+		<edge from-layer="1198" from-port="2" to-layer="1200" to-port="0" />
 		<edge from-layer="1199" from-port="0" to-layer="1200" to-port="1" />
-		<edge from-layer="1200" from-port="2" to-layer="1201" to-port="1" />
-		<edge from-layer="1201" from-port="2" to-layer="1203" to-port="0" />
-		<edge from-layer="1202" from-port="0" to-layer="1203" to-port="1" />
-		<edge from-layer="1203" from-port="2" to-layer="1208" to-port="0" />
-		<edge from-layer="1204" from-port="0" to-layer="1208" to-port="1" />
-		<edge from-layer="1205" from-port="0" to-layer="1208" to-port="2" />
-		<edge from-layer="1206" from-port="0" to-layer="1208" to-port="3" />
-		<edge from-layer="1207" from-port="0" to-layer="1208" to-port="4" />
-		<edge from-layer="1208" from-port="5" to-layer="1210" to-port="0" />
-		<edge from-layer="1209" from-port="0" to-layer="1210" to-port="1" />
-		<edge from-layer="1210" from-port="2" to-layer="1212" to-port="0" />
-		<edge from-layer="1211" from-port="0" to-layer="1212" to-port="1" />
-		<edge from-layer="1212" from-port="2" to-layer="1240" to-port="0" />
-		<edge from-layer="1213" from-port="0" to-layer="1214" to-port="0" />
-		<edge from-layer="1214" from-port="1" to-layer="1216" to-port="0" />
-		<edge from-layer="1215" from-port="0" to-layer="1216" to-port="1" />
-		<edge from-layer="1216" from-port="2" to-layer="1217" to-port="1" />
-		<edge from-layer="1217" from-port="2" to-layer="1219" to-port="0" />
-		<edge from-layer="1218" from-port="0" to-layer="1219" to-port="1" />
-		<edge from-layer="1219" from-port="2" to-layer="1224" to-port="0" />
-		<edge from-layer="1220" from-port="0" to-layer="1224" to-port="1" />
-		<edge from-layer="1221" from-port="0" to-layer="1224" to-port="2" />
-		<edge from-layer="1222" from-port="0" to-layer="1224" to-port="3" />
-		<edge from-layer="1223" from-port="0" to-layer="1224" to-port="4" />
-		<edge from-layer="1224" from-port="5" to-layer="1226" to-port="0" />
-		<edge from-layer="1225" from-port="0" to-layer="1226" to-port="1" />
-		<edge from-layer="1226" from-port="2" to-layer="1228" to-port="0" />
-		<edge from-layer="1227" from-port="0" to-layer="1228" to-port="1" />
-		<edge from-layer="1228" from-port="2" to-layer="1240" to-port="1" />
-		<edge from-layer="1229" from-port="0" to-layer="1230" to-port="0" />
-		<edge from-layer="1230" from-port="1" to-layer="1232" to-port="0" />
+		<edge from-layer="1200" from-port="2" to-layer="1202" to-port="0" />
+		<edge from-layer="1201" from-port="0" to-layer="1202" to-port="1" />
+		<edge from-layer="1202" from-port="2" to-layer="1203" to-port="2" />
+		<edge from-layer="1203" from-port="4" to-layer="1205" to-port="0" />
+		<edge from-layer="1204" from-port="0" to-layer="1205" to-port="1" />
+		<edge from-layer="1205" from-port="2" to-layer="1207" to-port="0" />
+		<edge from-layer="1206" from-port="0" to-layer="1207" to-port="1" />
+		<edge from-layer="1207" from-port="2" to-layer="1216" to-port="0" />
+		<edge from-layer="1208" from-port="0" to-layer="1209" to-port="0" />
+		<edge from-layer="1209" from-port="1" to-layer="1212" to-port="0" />
+		<edge from-layer="1210" from-port="0" to-layer="1211" to-port="0" />
+		<edge from-layer="1211" from-port="1" to-layer="1212" to-port="1" />
+		<edge from-layer="1212" from-port="2" to-layer="1214" to-port="0" />
+		<edge from-layer="1213" from-port="0" to-layer="1214" to-port="1" />
+		<edge from-layer="1214" from-port="2" to-layer="1215" to-port="0" />
+		<edge from-layer="1215" from-port="1" to-layer="1216" to-port="1" />
+		<edge from-layer="1216" from-port="2" to-layer="1218" to-port="0" />
+		<edge from-layer="1217" from-port="0" to-layer="1218" to-port="1" />
+		<edge from-layer="1218" from-port="2" to-layer="1219" to-port="0" />
+		<edge from-layer="1219" from-port="2" to-layer="1221" to-port="0" />
+		<edge from-layer="1220" from-port="0" to-layer="1221" to-port="1" />
+		<edge from-layer="1221" from-port="2" to-layer="1223" to-port="0" />
+		<edge from-layer="1222" from-port="0" to-layer="1223" to-port="1" />
+		<edge from-layer="1223" from-port="2" to-layer="1225" to-port="0" />
+		<edge from-layer="1224" from-port="0" to-layer="1225" to-port="1" />
+		<edge from-layer="1225" from-port="2" to-layer="1249" to-port="1" />
+		<edge from-layer="1225" from-port="2" to-layer="1234" to-port="0" />
+		<edge from-layer="1226" from-port="0" to-layer="1227" to-port="0" />
+		<edge from-layer="1227" from-port="1" to-layer="1230" to-port="0" />
+		<edge from-layer="1228" from-port="0" to-layer="1229" to-port="0" />
+		<edge from-layer="1229" from-port="1" to-layer="1230" to-port="1" />
+		<edge from-layer="1230" from-port="2" to-layer="1232" to-port="0" />
 		<edge from-layer="1231" from-port="0" to-layer="1232" to-port="1" />
-		<edge from-layer="1232" from-port="2" to-layer="1233" to-port="1" />
-		<edge from-layer="1233" from-port="2" to-layer="1235" to-port="0" />
-		<edge from-layer="1234" from-port="0" to-layer="1235" to-port="1" />
-		<edge from-layer="1235" from-port="2" to-layer="1237" to-port="0" />
-		<edge from-layer="1236" from-port="0" to-layer="1237" to-port="1" />
-		<edge from-layer="1237" from-port="2" to-layer="1239" to-port="0" />
-		<edge from-layer="1238" from-port="0" to-layer="1239" to-port="1" />
-		<edge from-layer="1239" from-port="2" to-layer="1240" to-port="2" />
-		<edge from-layer="1240" from-port="4" to-layer="1242" to-port="0" />
-		<edge from-layer="1241" from-port="0" to-layer="1242" to-port="1" />
+		<edge from-layer="1232" from-port="2" to-layer="1233" to-port="0" />
+		<edge from-layer="1233" from-port="1" to-layer="1234" to-port="1" />
+		<edge from-layer="1234" from-port="2" to-layer="1236" to-port="0" />
+		<edge from-layer="1235" from-port="0" to-layer="1236" to-port="1" />
+		<edge from-layer="1236" from-port="2" to-layer="1237" to-port="0" />
+		<edge from-layer="1237" from-port="1" to-layer="1246" to-port="0" />
+		<edge from-layer="1238" from-port="0" to-layer="1239" to-port="0" />
+		<edge from-layer="1239" from-port="1" to-layer="1242" to-port="0" />
+		<edge from-layer="1240" from-port="0" to-layer="1241" to-port="0" />
+		<edge from-layer="1241" from-port="1" to-layer="1242" to-port="1" />
 		<edge from-layer="1242" from-port="2" to-layer="1244" to-port="0" />
 		<edge from-layer="1243" from-port="0" to-layer="1244" to-port="1" />
-		<edge from-layer="1244" from-port="2" to-layer="1246" to-port="0" />
-		<edge from-layer="1245" from-port="0" to-layer="1246" to-port="1" />
-		<edge from-layer="1246" from-port="2" to-layer="1251" to-port="0" />
-		<edge from-layer="1247" from-port="0" to-layer="1251" to-port="1" />
-		<edge from-layer="1248" from-port="0" to-layer="1251" to-port="2" />
-		<edge from-layer="1249" from-port="0" to-layer="1251" to-port="3" />
-		<edge from-layer="1250" from-port="0" to-layer="1251" to-port="4" />
-		<edge from-layer="1251" from-port="5" to-layer="1256" to-port="0" />
-		<edge from-layer="1252" from-port="0" to-layer="1253" to-port="0" />
-		<edge from-layer="1253" from-port="1" to-layer="1255" to-port="0" />
+		<edge from-layer="1244" from-port="2" to-layer="1245" to-port="0" />
+		<edge from-layer="1245" from-port="1" to-layer="1246" to-port="1" />
+		<edge from-layer="1246" from-port="2" to-layer="1248" to-port="0" />
+		<edge from-layer="1247" from-port="0" to-layer="1248" to-port="1" />
+		<edge from-layer="1248" from-port="2" to-layer="1249" to-port="0" />
+		<edge from-layer="1249" from-port="2" to-layer="1251" to-port="0" />
+		<edge from-layer="1250" from-port="0" to-layer="1251" to-port="1" />
+		<edge from-layer="1251" from-port="2" to-layer="1253" to-port="0" />
+		<edge from-layer="1252" from-port="0" to-layer="1253" to-port="1" />
+		<edge from-layer="1253" from-port="2" to-layer="1255" to-port="0" />
 		<edge from-layer="1254" from-port="0" to-layer="1255" to-port="1" />
-		<edge from-layer="1255" from-port="2" to-layer="1256" to-port="1" />
-		<edge from-layer="1256" from-port="2" to-layer="1258" to-port="0" />
-		<edge from-layer="1257" from-port="0" to-layer="1258" to-port="1" />
-		<edge from-layer="1258" from-port="2" to-layer="1259" to-port="0" />
-		<edge from-layer="1259" from-port="2" to-layer="1261" to-port="0" />
-		<edge from-layer="1260" from-port="0" to-layer="1261" to-port="1" />
-		<edge from-layer="1261" from-port="2" to-layer="1263" to-port="0" />
-		<edge from-layer="1262" from-port="0" to-layer="1263" to-port="1" />
+		<edge from-layer="1255" from-port="2" to-layer="1258" to-port="0" />
+		<edge from-layer="1256" from-port="0" to-layer="1258" to-port="1" />
+		<edge from-layer="1257" from-port="0" to-layer="1258" to-port="2" />
+		<edge from-layer="1258" from-port="3" to-layer="1267" to-port="0" />
+		<edge from-layer="1259" from-port="0" to-layer="1260" to-port="0" />
+		<edge from-layer="1260" from-port="1" to-layer="1263" to-port="0" />
+		<edge from-layer="1261" from-port="0" to-layer="1262" to-port="0" />
+		<edge from-layer="1262" from-port="1" to-layer="1263" to-port="1" />
 		<edge from-layer="1263" from-port="2" to-layer="1265" to-port="0" />
 		<edge from-layer="1264" from-port="0" to-layer="1265" to-port="1" />
-		<edge from-layer="1265" from-port="2" to-layer="1267" to-port="0" />
-		<edge from-layer="1265" from-port="2" to-layer="1295" to-port="1" />
-		<edge from-layer="1266" from-port="0" to-layer="1267" to-port="1" />
-		<edge from-layer="1267" from-port="2" to-layer="1272" to-port="0" />
-		<edge from-layer="1268" from-port="0" to-layer="1272" to-port="1" />
-		<edge from-layer="1269" from-port="0" to-layer="1272" to-port="2" />
-		<edge from-layer="1270" from-port="0" to-layer="1272" to-port="3" />
-		<edge from-layer="1271" from-port="0" to-layer="1272" to-port="4" />
-		<edge from-layer="1272" from-port="5" to-layer="1277" to-port="0" />
+		<edge from-layer="1265" from-port="2" to-layer="1266" to-port="0" />
+		<edge from-layer="1266" from-port="1" to-layer="1267" to-port="1" />
+		<edge from-layer="1267" from-port="2" to-layer="1269" to-port="0" />
+		<edge from-layer="1268" from-port="0" to-layer="1269" to-port="1" />
+		<edge from-layer="1269" from-port="2" to-layer="1270" to-port="0" />
+		<edge from-layer="1270" from-port="1" to-layer="1279" to-port="0" />
+		<edge from-layer="1271" from-port="0" to-layer="1272" to-port="0" />
+		<edge from-layer="1272" from-port="1" to-layer="1275" to-port="0" />
 		<edge from-layer="1273" from-port="0" to-layer="1274" to-port="0" />
-		<edge from-layer="1274" from-port="1" to-layer="1276" to-port="0" />
-		<edge from-layer="1275" from-port="0" to-layer="1276" to-port="1" />
-		<edge from-layer="1276" from-port="2" to-layer="1277" to-port="1" />
-		<edge from-layer="1277" from-port="2" to-layer="1279" to-port="0" />
-		<edge from-layer="1278" from-port="0" to-layer="1279" to-port="1" />
-		<edge from-layer="1279" from-port="2" to-layer="1280" to-port="0" />
-		<edge from-layer="1280" from-port="1" to-layer="1282" to-port="0" />
-		<edge from-layer="1281" from-port="0" to-layer="1282" to-port="1" />
-		<edge from-layer="1282" from-port="2" to-layer="1287" to-port="0" />
-		<edge from-layer="1283" from-port="0" to-layer="1287" to-port="1" />
-		<edge from-layer="1284" from-port="0" to-layer="1287" to-port="2" />
-		<edge from-layer="1285" from-port="0" to-layer="1287" to-port="3" />
-		<edge from-layer="1286" from-port="0" to-layer="1287" to-port="4" />
-		<edge from-layer="1287" from-port="5" to-layer="1292" to-port="0" />
-		<edge from-layer="1288" from-port="0" to-layer="1289" to-port="0" />
-		<edge from-layer="1289" from-port="1" to-layer="1291" to-port="0" />
-		<edge from-layer="1290" from-port="0" to-layer="1291" to-port="1" />
-		<edge from-layer="1291" from-port="2" to-layer="1292" to-port="1" />
-		<edge from-layer="1292" from-port="2" to-layer="1294" to-port="0" />
-		<edge from-layer="1293" from-port="0" to-layer="1294" to-port="1" />
-		<edge from-layer="1294" from-port="2" to-layer="1295" to-port="0" />
-		<edge from-layer="1295" from-port="2" to-layer="1297" to-port="0" />
-		<edge from-layer="1296" from-port="0" to-layer="1297" to-port="1" />
-		<edge from-layer="1297" from-port="2" to-layer="1299" to-port="0" />
-		<edge from-layer="1298" from-port="0" to-layer="1299" to-port="1" />
-		<edge from-layer="1299" from-port="2" to-layer="1301" to-port="0" />
-		<edge from-layer="1300" from-port="0" to-layer="1301" to-port="1" />
-		<edge from-layer="1301" from-port="2" to-layer="1371" to-port="1" />
-		<edge from-layer="1301" from-port="2" to-layer="1303" to-port="0" />
-		<edge from-layer="1302" from-port="0" to-layer="1303" to-port="1" />
-		<edge from-layer="1303" from-port="2" to-layer="1308" to-port="0" />
-		<edge from-layer="1304" from-port="0" to-layer="1308" to-port="1" />
-		<edge from-layer="1305" from-port="0" to-layer="1308" to-port="2" />
-		<edge from-layer="1306" from-port="0" to-layer="1308" to-port="3" />
-		<edge from-layer="1307" from-port="0" to-layer="1308" to-port="4" />
-		<edge from-layer="1308" from-port="5" to-layer="1345" to-port="0" />
-		<edge from-layer="1308" from-port="5" to-layer="1329" to-port="0" />
-		<edge from-layer="1308" from-port="5" to-layer="1313" to-port="0" />
-		<edge from-layer="1309" from-port="0" to-layer="1310" to-port="0" />
-		<edge from-layer="1310" from-port="1" to-layer="1312" to-port="0" />
-		<edge from-layer="1311" from-port="0" to-layer="1312" to-port="1" />
-		<edge from-layer="1312" from-port="2" to-layer="1313" to-port="1" />
-		<edge from-layer="1313" from-port="2" to-layer="1315" to-port="0" />
-		<edge from-layer="1314" from-port="0" to-layer="1315" to-port="1" />
-		<edge from-layer="1315" from-port="2" to-layer="1320" to-port="0" />
-		<edge from-layer="1316" from-port="0" to-layer="1320" to-port="1" />
-		<edge from-layer="1317" from-port="0" to-layer="1320" to-port="2" />
-		<edge from-layer="1318" from-port="0" to-layer="1320" to-port="3" />
-		<edge from-layer="1319" from-port="0" to-layer="1320" to-port="4" />
-		<edge from-layer="1320" from-port="5" to-layer="1322" to-port="0" />
-		<edge from-layer="1321" from-port="0" to-layer="1322" to-port="1" />
-		<edge from-layer="1322" from-port="2" to-layer="1324" to-port="0" />
-		<edge from-layer="1323" from-port="0" to-layer="1324" to-port="1" />
-		<edge from-layer="1324" from-port="2" to-layer="1352" to-port="0" />
-		<edge from-layer="1325" from-port="0" to-layer="1326" to-port="0" />
-		<edge from-layer="1326" from-port="1" to-layer="1328" to-port="0" />
-		<edge from-layer="1327" from-port="0" to-layer="1328" to-port="1" />
-		<edge from-layer="1328" from-port="2" to-layer="1329" to-port="1" />
-		<edge from-layer="1329" from-port="2" to-layer="1331" to-port="0" />
-		<edge from-layer="1330" from-port="0" to-layer="1331" to-port="1" />
-		<edge from-layer="1331" from-port="2" to-layer="1336" to-port="0" />
-		<edge from-layer="1332" from-port="0" to-layer="1336" to-port="1" />
-		<edge from-layer="1333" from-port="0" to-layer="1336" to-port="2" />
-		<edge from-layer="1334" from-port="0" to-layer="1336" to-port="3" />
-		<edge from-layer="1335" from-port="0" to-layer="1336" to-port="4" />
-		<edge from-layer="1336" from-port="5" to-layer="1338" to-port="0" />
-		<edge from-layer="1337" from-port="0" to-layer="1338" to-port="1" />
-		<edge from-layer="1338" from-port="2" to-layer="1340" to-port="0" />
-		<edge from-layer="1339" from-port="0" to-layer="1340" to-port="1" />
-		<edge from-layer="1340" from-port="2" to-layer="1352" to-port="1" />
-		<edge from-layer="1341" from-port="0" to-layer="1342" to-port="0" />
-		<edge from-layer="1342" from-port="1" to-layer="1344" to-port="0" />
-		<edge from-layer="1343" from-port="0" to-layer="1344" to-port="1" />
-		<edge from-layer="1344" from-port="2" to-layer="1345" to-port="1" />
-		<edge from-layer="1345" from-port="2" to-layer="1347" to-port="0" />
-		<edge from-layer="1346" from-port="0" to-layer="1347" to-port="1" />
-		<edge from-layer="1347" from-port="2" to-layer="1349" to-port="0" />
-		<edge from-layer="1348" from-port="0" to-layer="1349" to-port="1" />
-		<edge from-layer="1349" from-port="2" to-layer="1351" to-port="0" />
-		<edge from-layer="1350" from-port="0" to-layer="1351" to-port="1" />
-		<edge from-layer="1351" from-port="2" to-layer="1352" to-port="2" />
-		<edge from-layer="1352" from-port="4" to-layer="1354" to-port="0" />
-		<edge from-layer="1353" from-port="0" to-layer="1354" to-port="1" />
-		<edge from-layer="1354" from-port="2" to-layer="1356" to-port="0" />
-		<edge from-layer="1355" from-port="0" to-layer="1356" to-port="1" />
-		<edge from-layer="1356" from-port="2" to-layer="1358" to-port="0" />
-		<edge from-layer="1357" from-port="0" to-layer="1358" to-port="1" />
-		<edge from-layer="1358" from-port="2" to-layer="1363" to-port="0" />
-		<edge from-layer="1359" from-port="0" to-layer="1363" to-port="1" />
-		<edge from-layer="1360" from-port="0" to-layer="1363" to-port="2" />
-		<edge from-layer="1361" from-port="0" to-layer="1363" to-port="3" />
-		<edge from-layer="1362" from-port="0" to-layer="1363" to-port="4" />
-		<edge from-layer="1363" from-port="5" to-layer="1368" to-port="0" />
-		<edge from-layer="1364" from-port="0" to-layer="1365" to-port="0" />
-		<edge from-layer="1365" from-port="1" to-layer="1367" to-port="0" />
-		<edge from-layer="1366" from-port="0" to-layer="1367" to-port="1" />
-		<edge from-layer="1367" from-port="2" to-layer="1368" to-port="1" />
-		<edge from-layer="1368" from-port="2" to-layer="1370" to-port="0" />
-		<edge from-layer="1369" from-port="0" to-layer="1370" to-port="1" />
-		<edge from-layer="1370" from-port="2" to-layer="1371" to-port="0" />
-		<edge from-layer="1371" from-port="2" to-layer="1373" to-port="0" />
-		<edge from-layer="1372" from-port="0" to-layer="1373" to-port="1" />
-		<edge from-layer="1373" from-port="2" to-layer="1375" to-port="0" />
-		<edge from-layer="1374" from-port="0" to-layer="1375" to-port="1" />
-		<edge from-layer="1375" from-port="2" to-layer="1377" to-port="0" />
-		<edge from-layer="1376" from-port="0" to-layer="1377" to-port="1" />
-		<edge from-layer="1377" from-port="2" to-layer="1379" to-port="0" />
-		<edge from-layer="1377" from-port="2" to-layer="1407" to-port="1" />
-		<edge from-layer="1378" from-port="0" to-layer="1379" to-port="1" />
-		<edge from-layer="1379" from-port="2" to-layer="1384" to-port="0" />
-		<edge from-layer="1380" from-port="0" to-layer="1384" to-port="1" />
-		<edge from-layer="1381" from-port="0" to-layer="1384" to-port="2" />
-		<edge from-layer="1382" from-port="0" to-layer="1384" to-port="3" />
-		<edge from-layer="1383" from-port="0" to-layer="1384" to-port="4" />
-		<edge from-layer="1384" from-port="5" to-layer="1389" to-port="0" />
-		<edge from-layer="1385" from-port="0" to-layer="1386" to-port="0" />
-		<edge from-layer="1386" from-port="1" to-layer="1388" to-port="0" />
-		<edge from-layer="1387" from-port="0" to-layer="1388" to-port="1" />
-		<edge from-layer="1388" from-port="2" to-layer="1389" to-port="1" />
-		<edge from-layer="1389" from-port="2" to-layer="1391" to-port="0" />
-		<edge from-layer="1390" from-port="0" to-layer="1391" to-port="1" />
-		<edge from-layer="1391" from-port="2" to-layer="1392" to-port="0" />
-		<edge from-layer="1392" from-port="1" to-layer="1394" to-port="0" />
-		<edge from-layer="1393" from-port="0" to-layer="1394" to-port="1" />
-		<edge from-layer="1394" from-port="2" to-layer="1399" to-port="0" />
-		<edge from-layer="1395" from-port="0" to-layer="1399" to-port="1" />
-		<edge from-layer="1396" from-port="0" to-layer="1399" to-port="2" />
-		<edge from-layer="1397" from-port="0" to-layer="1399" to-port="3" />
-		<edge from-layer="1398" from-port="0" to-layer="1399" to-port="4" />
-		<edge from-layer="1399" from-port="5" to-layer="1404" to-port="0" />
-		<edge from-layer="1400" from-port="0" to-layer="1401" to-port="0" />
-		<edge from-layer="1401" from-port="1" to-layer="1403" to-port="0" />
-		<edge from-layer="1402" from-port="0" to-layer="1403" to-port="1" />
-		<edge from-layer="1403" from-port="2" to-layer="1404" to-port="1" />
-		<edge from-layer="1404" from-port="2" to-layer="1406" to-port="0" />
-		<edge from-layer="1405" from-port="0" to-layer="1406" to-port="1" />
-		<edge from-layer="1406" from-port="2" to-layer="1407" to-port="0" />
-		<edge from-layer="1407" from-port="2" to-layer="1409" to-port="0" />
-		<edge from-layer="1408" from-port="0" to-layer="1409" to-port="1" />
-		<edge from-layer="1409" from-port="2" to-layer="1411" to-port="0" />
-		<edge from-layer="1410" from-port="0" to-layer="1411" to-port="1" />
-		<edge from-layer="1411" from-port="2" to-layer="1413" to-port="0" />
-		<edge from-layer="1412" from-port="0" to-layer="1413" to-port="1" />
-		<edge from-layer="1413" from-port="2" to-layer="1416" to-port="0" />
-		<edge from-layer="1414" from-port="0" to-layer="1416" to-port="1" />
-		<edge from-layer="1415" from-port="0" to-layer="1416" to-port="2" />
-		<edge from-layer="1416" from-port="3" to-layer="1418" to-port="0" />
-		<edge from-layer="1417" from-port="0" to-layer="1418" to-port="1" />
-		<edge from-layer="1418" from-port="2" to-layer="1423" to-port="0" />
-		<edge from-layer="1419" from-port="0" to-layer="1423" to-port="1" />
-		<edge from-layer="1420" from-port="0" to-layer="1423" to-port="2" />
-		<edge from-layer="1421" from-port="0" to-layer="1423" to-port="3" />
-		<edge from-layer="1422" from-port="0" to-layer="1423" to-port="4" />
-		<edge from-layer="1423" from-port="5" to-layer="1428" to-port="0" />
-		<edge from-layer="1424" from-port="0" to-layer="1425" to-port="0" />
-		<edge from-layer="1425" from-port="1" to-layer="1427" to-port="0" />
-		<edge from-layer="1426" from-port="0" to-layer="1427" to-port="1" />
-		<edge from-layer="1427" from-port="2" to-layer="1428" to-port="1" />
-		<edge from-layer="1428" from-port="2" to-layer="1430" to-port="0" />
-		<edge from-layer="1429" from-port="0" to-layer="1430" to-port="1" />
-		<edge from-layer="1430" from-port="2" to-layer="1431" to-port="0" />
-		<edge from-layer="1431" from-port="1" to-layer="1433" to-port="0" />
-		<edge from-layer="1432" from-port="0" to-layer="1433" to-port="1" />
-		<edge from-layer="1433" from-port="2" to-layer="1438" to-port="0" />
-		<edge from-layer="1434" from-port="0" to-layer="1438" to-port="1" />
-		<edge from-layer="1435" from-port="0" to-layer="1438" to-port="2" />
-		<edge from-layer="1436" from-port="0" to-layer="1438" to-port="3" />
-		<edge from-layer="1437" from-port="0" to-layer="1438" to-port="4" />
-		<edge from-layer="1438" from-port="5" to-layer="1443" to-port="0" />
-		<edge from-layer="1439" from-port="0" to-layer="1440" to-port="0" />
-		<edge from-layer="1440" from-port="1" to-layer="1442" to-port="0" />
-		<edge from-layer="1441" from-port="0" to-layer="1442" to-port="1" />
-		<edge from-layer="1442" from-port="2" to-layer="1443" to-port="1" />
-		<edge from-layer="1443" from-port="2" to-layer="1445" to-port="0" />
-		<edge from-layer="1444" from-port="0" to-layer="1445" to-port="1" />
-		<edge from-layer="1445" from-port="2" to-layer="1446" to-port="0" />
+		<edge from-layer="1274" from-port="1" to-layer="1275" to-port="1" />
+		<edge from-layer="1275" from-port="2" to-layer="1277" to-port="0" />
+		<edge from-layer="1276" from-port="0" to-layer="1277" to-port="1" />
+		<edge from-layer="1277" from-port="2" to-layer="1278" to-port="0" />
+		<edge from-layer="1278" from-port="1" to-layer="1279" to-port="1" />
+		<edge from-layer="1279" from-port="2" to-layer="1281" to-port="0" />
+		<edge from-layer="1280" from-port="0" to-layer="1281" to-port="1" />
+		<edge from-layer="1281" from-port="2" to-layer="1282" to-port="0" />
 	</edges>
 	<rt_info>
-		<Runtime_version value="2025.1.0-18503-6fec06580ab-releases/2025/1" />
+		<Runtime_version value="2025.3.0-19807-44526285f24-releases/2025/3" />
 		<conversion_parameters>
 			<framework value="pytorch" />
 			<is_python_object value="True" />
 		</conversion_parameters>
 		<nncf>
 			<friendly_names_were_updated value="True" />
-			<quantization>
-				<advanced_parameters value="{'overflow_fix': 'disable', 'quantize_outputs': False, 'inplace_statistics': True, 'disable_channel_alignment': True, 'disable_bias_correction': False, 'batchwise_statistics': None, 'activations_quantization_params': None, 'weights_quantization_params': None, 'quantizer_propagation_rule': '3', 'activations_range_estimator_params': {'min': {'statistics_type': None, 'aggregator_type': None, 'clipping_value': None, 'quantile_outlier_prob': 0.0001}, 'max': {'statistics_type': None, 'aggregator_type': None, 'clipping_value': None, 'quantile_outlier_prob': 0.0001}}, 'weights_range_estimator_params': {'min': {'statistics_type': None, 'aggregator_type': None, 'clipping_value': None, 'quantile_outlier_prob': 0.0001}, 'max': {'statistics_type': None, 'aggregator_type': None, 'clipping_value': None, 'quantile_outlier_prob': 0.0001}}, 'bias_correction_params': {'apply_for_all_nodes': False, 'threshold': None}, 'smooth_quant_alphas': {'convolution': -1, 'matmul': 0.95}, 'smooth_quant_alpha': None, 'backend_params': {}}" />
-				<fast_bias_correction value="True" />
+			<version value="2.19.0.dev0+8ef5ecd0" />
+			<weight_compression>
+				<advanced_parameters value="{'statistics_path': None, 'lora_adapter_rank': 256, 'group_size_fallback_mode': 'ignore', 'min_adjusted_group_size': 16, 'awq_params': {'subset_size': 32, 'percent_to_apply': 0.002, 'alpha_min': 0.0, 'alpha_max': 1.0, 'steps': 100, 'prefer_data_aware_scaling': True}, 'scale_estimation_params': {'subset_size': 64, 'initial_steps': 5, 'scale_steps': 5, 'weight_penalty': -1.0}, 'gptq_params': {'damp_percent': 0.1, 'block_size': 128, 'subset_size': 128}, 'lora_correction_params': {'adapter_rank': 8, 'num_iterations': 3, 'apply_regularization': True, 'subset_size': 128, 'use_int8_adapters': True}, 'backend_params': {}, 'codebook': None}" />
+				<all_layers value="False" />
+				<awq value="False" />
+				<backup_mode value="int8_asym" />
+				<compression_format value="dequantize" />
+				<gptq value="False" />
+				<group_size value="-1" />
 				<ignored_scope value="[]" />
-				<model_type value="transformer" />
-				<preset value="mixed" />
-				<subset_size value="300" />
-				<target_device value="ANY" />
-			</quantization>
+				<lora_correction value="False" />
+				<mode value="int8_asym" />
+				<ratio value="1.0" />
+				<scale_estimation value="False" />
+				<sensitivity_metric value="weight_quantization_error" />
+			</weight_compression>
 		</nncf>
 		<optimum>
-			<optimum_intel_version value="1.22.0" />
-			<optimum_version value="1.24.0" />
-			<pytorch_version value="2.7.0" />
-			<transformers_version value="4.48.3" />
+			<nncf_version value="2.19.0.dev0+8ef5ecd0" />
+			<optimum_intel_version value="1.26.0.dev0+9b51e6a" />
+			<optimum_version value="2.0.0" />
+			<pytorch_version value="2.8.0+cpu" />
+			<transformers_version value="4.55.4" />
 		</optimum>
 	</rt_info>
 </net>