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

[AMD] Fix DLPack Error in Aiter flydsl GEMM by Detaching MoE Gate Weight (#22262)

Browse Source 
Co-authored-by: bingxche <binxche@amd.com>
This commit is contained in:
Bingxu Chen
2026-04-07 23:42:10 -07:00
committed by GitHub
co-authored by bingxche
parent cd373667cd
commit de0cfed159
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/rocm_linear_utils.py
+1 -1
View File
@@ -11,7 +11,7 @@ def aiter_dsv3_router_gemm(
weight: torch.Tensor, weight: torch.Tensor,
): ):
"""Use aiter tuned GEMM dispatcher (tgemm.mm) to automatically select the GEMM kernel.""" """Use aiter tuned GEMM dispatcher (tgemm.mm) to automatically select the GEMM kernel."""
return tgemm.mm(hidden_states, weight, otype=hidden_states.dtype) return tgemm.mm(hidden_states, weight.detach(), otype=hidden_states.dtype)
def get_dsv3_gemm_output_zero_allocator_size( def get_dsv3_gemm_output_zero_allocator_size(