[Bugfix] Use native SiLU activation in CPU fused MoE (#45961)
Signed-off-by: Alden Lobo <alden.lobo@arm.com> Co-authored-by: Alden Lobo <alden.lobo@arm.com>
Alden Lobo committed
Jun 29, 2026 at 04:12 UTC
e1861078704b0b091206e83cdd64eaf10b1967ef
1 file changed
+1
-1
vllm/model_executor/layers/fused_moe/cpu_fused_moe.py
+1
-1
@@ -51,7 +51,7 @@ def _gelu_and_mul(
51
# Uses static methods or standalone functions to avoid instantiating CustomOp
52
# classes, which would call get_current_vllm_config() before config is set.
53
_CPU_MOE_ACT_FN: dict[MoEActivation, Callable[[torch.Tensor], torch.Tensor]] = {
54
- MoEActivation.SILU: lambda x: SiluAndMul(compile_native=False).forward_native(x),
54
+ MoEActivation.SILU: SiluAndMul.forward_native,
55
MoEActivation.SWIGLUOAI: _swigluoai_forward_native,
56
MoEActivation.GELU: _gelu_and_mul,
57
MoEActivation.GELU_TANH: (