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 7, 11:30 UTC
[XPU] fix mxfp4 activation type (#35691)
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Kunshang Ji committed
Mar 2, 2026 at 11:48 UTC
92f5d0f070ec9e0ca5fe72af672138a9bbd1cb68
1 file changed
+1
-1
vllm/model_executor/layers/quantization/mxfp4.py
+1
-1
@@ -1258,7 +1258,7 @@ class XpuMxfp4MoEMethod(Mxfp4MoEMethod):
1258
topk_weights=routing_weights,
1259
topk_ids=selected_experts,
1260
n_experts_per_token=layer.top_k,
1261
- activation=layer.activation,
1261
+ activation=layer.activation.value,
1262
num_experts=layer.local_num_experts,
1263
is_mxfp4=True,
1264
)