Create quantize_config.json
Browse files- quantize_config.json +22 -0
quantize_config.json
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bits": 4,
|
| 3 |
+
"block_name_to_quantize": "transformer.h",
|
| 4 |
+
"damp_percent": 0.1,
|
| 5 |
+
"dataset": "c4",
|
| 6 |
+
"desc_act": false,
|
| 7 |
+
"disable_exllama": true,
|
| 8 |
+
"group_size": 128,
|
| 9 |
+
"max_input_length": null,
|
| 10 |
+
"model_seqlen": 2048,
|
| 11 |
+
"module_name_preceding_first_block": [
|
| 12 |
+
"transformer.wte",
|
| 13 |
+
"transformer.wpe",
|
| 14 |
+
"transformer.drop"
|
| 15 |
+
],
|
| 16 |
+
"pad_token_id": null,
|
| 17 |
+
"quant_method": "gptq",
|
| 18 |
+
"sym": true,
|
| 19 |
+
"tokenizer": null,
|
| 20 |
+
"true_sequential": true,
|
| 21 |
+
"use_cuda_fp16": true
|
| 22 |
+
}
|