[Kernel] Add cuda kernel for gpt_oss activation (#22538)

Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
2025-08-15 08:21:29 +08:00
parent 39cd09dc86
commit 81f4b96481
8 changed files with 150 additions and 24 deletions
--- a/vllm/model_executor/layers/quantization/utils/mxfp4_utils.py
+++ b/vllm/model_executor/layers/quantization/utils/mxfp4_utils.py
@@ -68,7 +68,7 @@ def _can_support_mxfp4(use_grouped_topk: bool = False,
    return not (use_grouped_topk or topk_group or num_expert_group
                or expert_map or custom_routing_function
                or e_score_correction_bias or apply_router_weight_on_input
-                or scoring_func != "softmax" or activation != "swiglu_oai"
+                or scoring_func != "softmax" or activation != "swigluoai"
                or expert_load_view or logical_to_physical_map
                or logical_replica_count)