From 8c1d4acbfe70a1dffdb9a3db57b4d12329350295 Mon Sep 17 00:00:00 2001 From: "Li, Jiang" Date: Fri, 19 Sep 2025 15:27:22 +0800 Subject: [PATCH] [CPU] Disable oneDNN linear on non-x86 platforms (#25166) Signed-off-by: jiang1.li --- vllm/model_executor/layers/utils.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/vllm/model_executor/layers/utils.py b/vllm/model_executor/layers/utils.py index d2b135c1e..a1675ffba 100644 --- a/vllm/model_executor/layers/utils.py +++ b/vllm/model_executor/layers/utils.py @@ -7,7 +7,7 @@ import torch from vllm import _custom_ops as ops from vllm import envs -from vllm.platforms import current_platform +from vllm.platforms import CpuArchEnum, current_platform from vllm.utils import direct_register_custom_op @@ -167,7 +167,8 @@ def dispatch_cpu_unquantized_gemm( if remove_weight: layer.weight = torch.nn.Parameter(torch.empty(0), requires_grad=False) - elif ops._supports_onednn: + elif (ops._supports_onednn + and current_platform.get_cpu_architecture() == CpuArchEnum.X86): origin_weight = layer.weight if remove_weight: layer.weight = torch.nn.Parameter(torch.empty(0),