Restruct sgl-kernel benchmark (#10861)

2025-09-25 07:45:25 +08:00
parent 7a06ef984d
commit c4e314f986
27 changed files with 425 additions and 319 deletions
--- a/sgl-kernel/benchmark/bench_per_token_quant_fp8.py
+++ b/sgl-kernel/benchmark/bench_per_token_quant_fp8.py
@@ -160,7 +160,7 @@ def benchmark_quantization(batch_size, seq_len, hidden_dim, provider):
    elif provider == "sglang":
        fn = lambda: sglang_per_token_quant_fp8(x.clone())

-    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