minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

fix: add bias when enable mm fallback variant (#17690)

Browse Source
This commit is contained in:
Yisheng Gong
2026-01-28 09:50:49 +08:00
committed by GitHub
parent 647428d8d6
commit 1c4616a034
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/batch_invariant_ops/batch_invariant_ops.py
+4 -1
View File
@@ -297,7 +297,10 @@ def matmul_persistent(
return _matmul_persistent_deepgemm(a=a, b=b, bias=bias) return _matmul_persistent_deepgemm(a=a, b=b, bias=bias)
if _ENABLE_MM_FALLBACK_VARIANT: if _ENABLE_MM_FALLBACK_VARIANT:
return torch.einsum("ik,kj->ij", a, b) out = torch.einsum("ik,kj->ij", a, b)
if bias is not None:
out += bias
return out
return _matmul_persistent_triton(a=a, b=b, bias=bias) return _matmul_persistent_triton(a=a, b=b, bias=bias)