EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[Doc] add doc for quantization w8a8_fp8 or w8a8_int8 (#4495)

Browse Source
This commit is contained in:
HandH1998
2025-03-17 17:25:00 +08:00
committed by GitHub
parent 91ba98fe50
commit f2ab37e500
1 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
docs/backend/quantization.md
View File

@@ -29,6 +29,15 @@ python3 -m sglang.launch_server \
--port 30000 --host 0.0.0.0 --port 30000 --host 0.0.0.0
``` ```
Take note, if your model is **per-channel quantized (INT8 or FP8) with per-token dynamic quantization activation**, you can opt to include `--quantization w8a8_int8` or `--quantization w8a8_fp8` to invoke the corresponding CUTLASS int8_kernel or fp8_kernel in sgl-kernel. This action will ignore the Hugging Face config's quantization settings. For instance, with `neuralmagic/Meta-Llama-3.1-8B-Instruct-FP8-dynamic`, if you execute with `--quantization w8a8_fp8`, the system will use the `W8A8Fp8Config` from SGLang to invoke the sgl-kernel, rather than the `CompressedTensorsConfig` for vLLM kernels.
```bash
python3 -m sglang.launch_server \
--model-path neuralmagic/Meta-Llama-3.1-8B-Instruct-FP8-dynamic \
--quantization w8a8_fp8 \
--port 30000 --host 0.0.0.0
```
### Examples of Offline Model Quantization ### Examples of Offline Model Quantization
#### Using [GPTQModel](https://github.com/ModelCloud/GPTQModel) #### Using [GPTQModel](https://github.com/ModelCloud/GPTQModel)