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

Restruct sgl-kernel benchmark (#10861)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2025-09-25 07:45:25 +08:00
committed by GitHub
parent 7a06ef984d
commit c4e314f986
27 changed files with 425 additions and 319 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
sgl-kernel/benchmark/bench_moe_topk_softmax.py
View File

@@ -97,7 +97,7 @@ def benchmark(num_tokens, num_experts, topk, provider):
fn = lambda: sglang_topk_softmax(gating_output, topk)
quantiles = [0.5, 0.2, 0.8]
ms, min_ms, max_ms = triton.testing.do_bench(fn, quantiles=quantiles)
ms, min_ms, max_ms = triton.testing.do_bench_cudagraph(fn, quantiles=quantiles)
return 1000 * ms, 1000 * max_ms, 1000 * min_ms