From 2ac5b983950843e02b46f6648ade184dcfe8f09f Mon Sep 17 00:00:00 2001 From: Vincent Zhong <207368749+vincentzed@users.noreply.github.com> Date: Fri, 5 Dec 2025 23:12:57 -0500 Subject: [PATCH] fix: fix rmsnorm -> layernorm in qwen3 omni (#11791) Co-authored-by: Brayden Zhong --- python/sglang/srt/models/qwen3_omni_moe.py | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/python/sglang/srt/models/qwen3_omni_moe.py b/python/sglang/srt/models/qwen3_omni_moe.py index 8663e5ac5..ae5b8332d 100644 --- a/python/sglang/srt/models/qwen3_omni_moe.py +++ b/python/sglang/srt/models/qwen3_omni_moe.py @@ -31,7 +31,6 @@ from sglang.srt.configs.qwen3_omni import ( ) from sglang.srt.configs.qwen3_vl import Qwen3VLMoeConfig from sglang.srt.layers.attention.vision import VisionAttention -from sglang.srt.layers.layernorm import RMSNorm from sglang.srt.layers.linear import ColumnParallelLinear, RowParallelLinear from sglang.srt.layers.moe.fused_moe_triton.layer import FusedMoE from sglang.srt.layers.quantization.base_config import QuantizationConfig @@ -318,7 +317,7 @@ class Qwen3OmniMoeVisionPatchMerger(nn.Module): super().__init__() self.hidden_size = context_dim * (spatial_merge_size**2) self.use_postshuffle_norm = use_postshuffle_norm - self.ln_q = RMSNorm( + self.ln_q = nn.LayerNorm( self.hidden_size if use_postshuffle_norm else context_dim, eps=1e-6 ) self.mlp = nn.ModuleList(