From 3339c8107292346b173bede6989bae060534cf11 Mon Sep 17 00:00:00 2001 From: gongwei-130 <56567052+gongwei-130@users.noreply.github.com> Date: Sat, 29 Nov 2025 12:17:41 -0800 Subject: [PATCH] fix RuntimeError: RMSNorm failed with error code an illegal memory access was encountered (#14135) --- python/sglang/srt/layers/moe/fused_moe_triton/layer.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/layer.py b/python/sglang/srt/layers/moe/fused_moe_triton/layer.py index 224ab3c7f..6e0487a1d 100644 --- a/python/sglang/srt/layers/moe/fused_moe_triton/layer.py +++ b/python/sglang/srt/layers/moe/fused_moe_triton/layer.py @@ -1117,7 +1117,7 @@ class FlashInferFP4MoE(FusedMoE): topk_config = topk_output.topk_config hs_fp4, hs_scale_linear = self._quantize_hidden_states_fp4(hidden_states) - + router_logits = router_logits.to(torch.float32) routing_method_type = self.routing_method_type assert ( routing_method_type is not None