From f873a80841464dec2994e136f203fdf6c9f66f30 Mon Sep 17 00:00:00 2001
From: narutolhy <582909902@qq.com>
Date: Sat, 21 Feb 2026 16:42:00 -0800
Subject: [PATCH 1/2] qwen3 vl skip layer id for pp

---
 python/sglang/srt/models/qwen3_vl_moe.py | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/python/sglang/srt/models/qwen3_vl_moe.py b/python/sglang/srt/models/qwen3_vl_moe.py
index f98460665d37..c1544e9b34f2 100644
--- a/python/sglang/srt/models/qwen3_vl_moe.py
+++ b/python/sglang/srt/models/qwen3_vl_moe.py
@@ -26,6 +26,7 @@
 from sglang.srt.eplb.expert_location import ModelConfigForExpertLocation
 from sglang.srt.layers.moe.fused_moe_triton.layer import FusedMoE
 from sglang.srt.layers.quantization.base_config import QuantizationConfig
+from sglang.srt.layers.utils import get_layer_id
 from sglang.srt.model_executor.forward_batch_info import ForwardBatch, PPProxyTensors
 from sglang.srt.model_loader.weight_utils import default_weight_loader
 from sglang.srt.models.qwen3_moe import Qwen3MoeDecoderLayer, Qwen3MoeModel
@@ -232,6 +233,16 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
         params_dict = self._cached_params_dict
         for name, loaded_weight in weights:
             name = name.replace(r"model.language_model.", r"model.")
+            layer_id = get_layer_id(name)
+            if (
+                layer_id is not None
+                and hasattr(self.model, "start_layer")
+                and (
+                    layer_id < self.model.start_layer
+                    or layer_id >= self.model.end_layer
+                )
+            ):
+                continue
 
             for param_name, weight_name, shard_id in stacked_params_mapping:
                 if "experts.gate_up_proj" in name or "experts.down_proj" in name:

From 27311a40cb1f72953a8780869d7b9f467a8cd317 Mon Sep 17 00:00:00 2001
From: narutolhy <582909902@qq.com>
Date: Sat, 21 Feb 2026 19:34:16 -0800
Subject: [PATCH 2/2] fix

---
 python/sglang/srt/models/qwen3_vl_moe.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/python/sglang/srt/models/qwen3_vl_moe.py b/python/sglang/srt/models/qwen3_vl_moe.py
index c1544e9b34f2..88f63dbcad72 100644
--- a/python/sglang/srt/models/qwen3_vl_moe.py
+++ b/python/sglang/srt/models/qwen3_vl_moe.py
@@ -235,7 +235,8 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
             name = name.replace(r"model.language_model.", r"model.")
             layer_id = get_layer_id(name)
             if (
-                layer_id is not None
+                "visual" not in name
+                and layer_id is not None
                 and hasattr(self.model, "start_layer")
                 and (
                     layer_id < self.model.start_layer