[Bugfix][MoE] Support GELU tanh in FlashInfer B12x MoE (#51819)
Signed-off-by: Andrii Skliar <askliar@nvidia.com> Co-authored-by: Andrii Skliar <askliar@nvidia.com> Co-authored-by: OpenAI Codex <codex@openai.com>
Andrii Skliar committed
Aug 11, 2026 at 17:07 UTC
0fb9897df12963e9bd0d58547fc997e0702d95ec
1 file changed
+6
-1
vllm/model_executor/layers/fused_moe/experts/flashinfer_b12x_moe.py
+6
-1
@@ -45,6 +45,7 @@ class FlashInferB12xExperts(mk.FusedMoEExpertsModular):
45
46
_ACTIVATION_MAP: dict[MoEActivation, str] = {
47
MoEActivation.SILU: "silu",
48
+ MoEActivation.GELU_TANH: "gelu_tanh",
49
MoEActivation.RELU2_NO_MUL: "relu2",
50
}
51
@@ -188,7 +189,11 @@ class FlashInferB12xExperts(mk.FusedMoEExpertsModular):
189
190
@staticmethod
191
def _supports_activation(activation: MoEActivation) -> bool:
191
- return activation in (MoEActivation.SILU, MoEActivation.RELU2_NO_MUL)
192
+ return activation in (
193
+ MoEActivation.SILU,
194
+ MoEActivation.GELU_TANH,
195
+ MoEActivation.RELU2_NO_MUL,
196
+ )
197
198
@staticmethod
199
def _supports_parallel_config(moe_parallel_config: FusedMoEParallelConfig) -> bool: