Add DeepSeek distilled versions of LLama 8B
Browse files
inference-cache-config/llama-variants.json
CHANGED
|
@@ -1,4 +1,48 @@
|
|
| 1 |
{
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
"princeton-nlp/Sheared-LLaMA-1.3B": [
|
| 3 |
{
|
| 4 |
"batch_size": 1,
|
|
|
|
| 1 |
{
|
| 2 |
+
"deepseek-ai/DeepSeek-R1-Distill-Llama-8B": [
|
| 3 |
+
{
|
| 4 |
+
"batch_size": 1,
|
| 5 |
+
"sequence_length": 4096,
|
| 6 |
+
"num_cores": 2,
|
| 7 |
+
"auto_cast_type": "bf16"
|
| 8 |
+
},
|
| 9 |
+
{
|
| 10 |
+
"batch_size": 4,
|
| 11 |
+
"sequence_length": 4096,
|
| 12 |
+
"num_cores": 2,
|
| 13 |
+
"auto_cast_type": "bf16"
|
| 14 |
+
},
|
| 15 |
+
{
|
| 16 |
+
"batch_size": 8,
|
| 17 |
+
"sequence_length": 4096,
|
| 18 |
+
"num_cores": 2,
|
| 19 |
+
"auto_cast_type": "bf16"
|
| 20 |
+
},
|
| 21 |
+
{
|
| 22 |
+
"batch_size": 4,
|
| 23 |
+
"sequence_length": 4096,
|
| 24 |
+
"num_cores": 8,
|
| 25 |
+
"auto_cast_type": "bf16"
|
| 26 |
+
},
|
| 27 |
+
{
|
| 28 |
+
"batch_size": 8,
|
| 29 |
+
"sequence_length": 4096,
|
| 30 |
+
"num_cores": 8,
|
| 31 |
+
"auto_cast_type": "bf16"
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"batch_size": 16,
|
| 35 |
+
"sequence_length": 4096,
|
| 36 |
+
"num_cores": 8,
|
| 37 |
+
"auto_cast_type": "bf16"
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"batch_size": 32,
|
| 41 |
+
"sequence_length": 4096,
|
| 42 |
+
"num_cores": 8,
|
| 43 |
+
"auto_cast_type": "bf16"
|
| 44 |
+
}
|
| 45 |
+
],
|
| 46 |
"princeton-nlp/Sheared-LLaMA-1.3B": [
|
| 47 |
{
|
| 48 |
"batch_size": 1,
|