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

add deepgemm and sglang fp8 block-wise gemm benchmark (#3893)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2025-03-02 15:01:58 +08:00
committed by GitHub
parent 407e2b923d
commit 90a55e2566
2 changed files with 320 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
benchmark/kernels/deepseek/README.md Normal file
View File

@@ -0,0 +1,6 @@
## DeepSeek kernels benchmark
- `benchmark_deepgemm_fp8_gemm.py`
- You should install [DeepGemm](https://github.com/deepseek-ai/DeepGEMM) from source before run `benchmark_deepgemm_fp8_gemm.py`.
- You can use the `--run_correctness` parameter to verify all kernels results's correctness.
- You can use the `--tp_size` parameter to benchmark all FP8 w8a8 block-wise matrix multiplications involved in DeepSeek V3/R1 under the current tensor parallelism (TP) setting. This benchmark compares DeepSeek's open-source [DeepGemm](https://github.com/deepseek-ai/DeepGEMM) implementation with SGLang's and VLLM Triton implementation.