Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -26,7 +26,8 @@ python3 quantize_quark.py \
|
|
| 26 |
--kv_cache_dtype fp8 \
|
| 27 |
--num_calib_data 128 \
|
| 28 |
--model_export quark_safetensors \
|
| 29 |
-
--no_weight_matrix_merge
|
|
|
|
| 30 |
|
| 31 |
# If model size is too large for single GPU, please use multi GPU instead.
|
| 32 |
python3 quantize_quark.py \
|
|
@@ -37,7 +38,8 @@ python3 quantize_quark.py \
|
|
| 37 |
--num_calib_data 128 \
|
| 38 |
--model_export quark_safetensors \
|
| 39 |
--no_weight_matrix_merge \
|
| 40 |
-
--multi_gpu
|
|
|
|
| 41 |
```
|
| 42 |
## Deployment
|
| 43 |
Quark has its own export format and allows FP8 quantized models to be efficiently deployed using the vLLM backend(vLLM-compatible).
|
|
|
|
| 26 |
--kv_cache_dtype fp8 \
|
| 27 |
--num_calib_data 128 \
|
| 28 |
--model_export quark_safetensors \
|
| 29 |
+
--no_weight_matrix_merge \
|
| 30 |
+
--custom_mode fp8
|
| 31 |
|
| 32 |
# If model size is too large for single GPU, please use multi GPU instead.
|
| 33 |
python3 quantize_quark.py \
|
|
|
|
| 38 |
--num_calib_data 128 \
|
| 39 |
--model_export quark_safetensors \
|
| 40 |
--no_weight_matrix_merge \
|
| 41 |
+
--multi_gpu \
|
| 42 |
+
--custom_mode fp8
|
| 43 |
```
|
| 44 |
## Deployment
|
| 45 |
Quark has its own export format and allows FP8 quantized models to be efficiently deployed using the vLLM backend(vLLM-compatible).
|