also fix weight init issue

modeling_deepseek.py

@@ -1244,7 +1244,7 @@ class DeepseekV3PreTrainedModel(PreTrainedModel):
 
     def _init_weights(self, module):
         std = self.config.initializer_range
-        if isinstance(module, nn.Linear):
+        if isinstance(module, nn.Linear) and hasattr(module, "weight"):
             module.weight.data.normal_(mean=0.0, std=std)
             if module.bias is not None:
                 module.bias.data.zero_()

modeling_kimi_k25.py

@@ -638,6 +638,20 @@ class MoonViT3dPretrainedModel(PreTrainedModel):
     _no_split_modules = ['PackingTransformer']
     _supports_flash_attn_2 = True
     _supports_sdpa = True
+    
+    def _init_weights(self, module):
+        # Default PreTrainedModel._init_weights treats nn.Linear subclasses (e.g.
+        # compressed_tensors.CompressedLinear) as Linear but those layers may have
+        # no `weight` after compression; skip in that case.
+        std = getattr(self.config, "initializer_range", 0.02)
+        if isinstance(module, (nn.Linear, nn.Conv2d)) and hasattr(module, "weight"):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
 
     def __init__(self, config, *inputs, **kwargs):
         super().__init__(config, *inputs, **kwargs)
@@ -786,7 +800,7 @@ class KimiK25PreTrainedModel(PreTrainedModel):
         if hasattr(module, "class_embedding"):
             module.class_embedding.data.normal_(mean=0.0, std=std)
 
-        if isinstance(module, (nn.Linear, nn.Conv2d)):
+        if isinstance(module, (nn.Linear, nn.Conv2d)) and hasattr(module, "weight"):
             module.weight.data.normal_(mean=0.0, std=std)
             if module.bias is not None:
                 module.bias.data.zero_()