xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-01-23 20:14:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ROCm] Fix for import when building with upstream triton for gfx1100 for gpt-oss serving (#29127)

Browse Source 
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
This commit is contained in:
Hongxia Yang 2025-11-20 22:30:07 -05:00 committed by GitHub
parent e1eefa4c40
commit 3f5f36da3f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/model_executor/layers/quantization/utils/mxfp4_utils.py
View File

@ -39,15 +39,15 @@ def _swizzle_mxfp4(quant_tensor, scale, num_warps):
value_layout = StridedLayout
scale_layout = StridedLayout
elif current_platform.is_rocm():
from triton_kernels.tensor_details.layout import (
GFX950MXScaleLayout,
StridedLayout,
)
from vllm.platforms.rocm import on_gfx950
value_layout = StridedLayout
scale_layout = GFX950MXScaleLayout if on_gfx950() else StridedLayout
if on_gfx950():
from triton_kernels.tensor_details.layout import GFX950MXScaleLayout
scale_layout = GFX950MXScaleLayout
else:
scale_layout = StridedLayout
else:
value_layout, value_layout_opts = layout.make_default_matmul_mxfp4_w_layout(
mx_axis=1