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

[Bug] Fix Issue#10215 (#10572)

Browse Source
This commit is contained in:
Yuhao Yao
2025-09-25 09:51:50 +08:00
committed by GitHub
parent c4e314f986
commit fe531d6f4e
2 changed files with 9 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
sgl-kernel/tests/test_cutlass_w4a8_moe_mm.py
View File

@@ -157,8 +157,8 @@ def _per_tensor_quant_fp8(
reason="cutlass_w4a8_moe_mm is only supported on sm90",
)
@pytest.mark.parametrize("batch_size", [2, 4, 8, 16, 32])
@pytest.mark.parametrize("k", [512, 1024, 2048, 4096, 7168])
@pytest.mark.parametrize("n", [256, 512, 1024, 2048])
@pytest.mark.parametrize("k", [256, 512, 1024, 2048, 4096, 7168])
@pytest.mark.parametrize("n", [256, 512, 1024, 2048, 7168])
@pytest.mark.parametrize("num_experts", [2, 4, 6, 8])
def test_int4_fp8_grouped_gemm_multi_experts(batch_size, k, n, num_experts):
torch.manual_seed(0)