biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Fix persistent_masked_m_silu_mul_quant tests (#28366)

Browse Source 
Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
This commit is contained in:
Varun Sundar Rabindranath
2025-11-10 12:21:52 -05:00
committed by GitHub
parent d0e186c16f
commit b039bfda8f
3 changed files with 16 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/kernels/moe/test_silu_mul_fp8_quant_deep_gemm.py
View File

@@ -25,6 +25,7 @@ CASES = [
(8, 16, 128 * 2, fp8_dtype),
(8, 16, 128 * 3, fp8_dtype),
(8, 64, 7168, fp8_dtype),
(8, 128, 128 * 33, fp8_dtype),
(8, 128, 7168, fp8_dtype),
(8, 512, 7168, fp8_dtype),
(8, 1024, 7168, fp8_dtype),
@@ -54,8 +55,10 @@ def test_silu_mul_fp8_quant_deep_gemm(E, T, H, fp8_type):
)
# Run the SiLU V2 kernel
# TODO (varun): use_e8m0 is set to false as the reference impl does
# not handle that case.
y_q, y_s = persistent_masked_m_silu_mul_quant(
y, tokens_per_expert, group_size=group_size
y, tokens_per_expert, group_size=group_size, use_ue8m0=False
)
torch.cuda.synchronize()