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

[AMD] Fix missing per_token_group_quant_fp8 for ROCm (#5140)

Browse Source
This commit is contained in:
Hubert Lu
2025-04-07 22:38:25 -07:00
committed by GitHub
parent 9731eca77b
commit afb752bcbe
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py
View File

@@ -765,6 +765,8 @@ def invoke_fused_moe_kernel(
from sglang.srt.layers.quantization.fp8_kernel import ( from sglang.srt.layers.quantization.fp8_kernel import (
sglang_per_token_group_quant_fp8, sglang_per_token_group_quant_fp8,
) )
else:
from sglang.srt.layers.quantization.fp8_kernel import per_token_group_quant_fp8
assert topk_weights.stride(1) == 1 assert topk_weights.stride(1) == 1
assert sorted_token_ids.stride(0) == 1 assert sorted_token_ids.stride(0) == 1