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

[Bugfix] Fix DSV3 kernels breaking _C and _moe_C on unsupported arches (#35123)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2026-02-23 20:11:27 -05:00
committed by GitHub
parent 22a97e6613
commit 3ef9fd0f98
5 changed files with 12 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
csrc/torch_bindings.cpp
View File

@@ -242,7 +242,7 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
// DeepSeek V3 fused A GEMM (SM 9.0+, bf16 only, 1-16 tokens).
ops.def(
"dsv3_fused_a_gemm(Tensor! output, Tensor mat_a, Tensor mat_b) -> ()");
ops.impl("dsv3_fused_a_gemm", torch::kCUDA, &dsv3_fused_a_gemm);
// conditionally compiled so impl registration is in source file
// Quantized GEMM for AWQ.
ops.def(