diff --git a/python/sglang/srt/layers/quantization/awq/awq.py b/python/sglang/srt/layers/quantization/awq/awq.py
index 0b63dcb58..3fdcc7399 100644
--- a/python/sglang/srt/layers/quantization/awq/awq.py
+++ b/python/sglang/srt/layers/quantization/awq/awq.py
@@ -194,6 +194,8 @@ class AWQCPUConfig(AWQConfig):
             layer.scheme = self.get_linear_scheme(layer)
             return AWQLinearMethod(self)
         elif isinstance(layer, FusedMoE):
+            if is_layer_skipped_awq(prefix, self.modules_to_not_convert):
+                return None
             layer.scheme = self.get_moe_scheme(layer)
             return AWQMoEMethod(self)
         return None
diff --git a/python/sglang/srt/models/qwen3_5.py b/python/sglang/srt/models/qwen3_5.py
index 682b7ca1e..fe26e906e 100644
--- a/python/sglang/srt/models/qwen3_5.py
+++ b/python/sglang/srt/models/qwen3_5.py
@@ -301,7 +301,7 @@ class Qwen3_5GatedDeltaNet(nn.Module):
             group_size=None,
             norm_before_gate=True,
             device=torch.get_device_module().current_device(),
-            dtype=config.torch_dtype,
+            dtype=torch.get_default_dtype(),
             **(
                 {"activation": self.output_gate_type}
                 if self.output_gate_type is not None