A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[gpt-oss] Fix mxfp4 support (#22700)
Signed-off-by: Chen Zhang <zhangch99@outlook.com>
Chen Zhang committed
Aug 11, 2025 at 21:22 UTC
bbaf9e9cb15af23e7a1fd250bf49a5efb15cadf7
1 file changed
+1
-1
vllm/model_executor/layers/quantization/utils/mxfp4_utils.py
+1
-1
@@ -68,7 +68,7 @@ def _can_support_mxfp4(use_grouped_topk: bool = False,
68
return not (use_grouped_topk or topk_group or num_expert_group
69
or expert_map or custom_routing_function
70
or e_score_correction_bias or apply_router_weight_on_input
71
- or scoring_func != "softmax" or activation != "silu"
71
+ or scoring_func != "softmax" or activation != "swiglu_oai"
72
or expert_load_view or logical_to_physical_map
73
or logical_replica_count)
74