update deepseek and qwen to skip weight_shape loading

vllm/model_executor/models/deepseek_v2.py

@@ -553,6 +553,8 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
 
         params_dict = dict(self.named_parameters())
         for name, loaded_weight in weights:
+            if "weight_shape" in name:
+                continue
             if "rotary_emb.inv_freq" in name:
                 continue
             for (param_name, weight_name, shard_id) in stacked_params_mapping:

vllm/model_executor/models/qwen2_moe.py

@@ -453,6 +453,8 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
 
         params_dict = dict(self.named_parameters())
         for name, loaded_weight in weights:
+            if "weight_shape" in name:
+                continue
             if "rotary_emb.inv_freq" in name:
                 continue
             for (param_name, weight_name, shard_id) in stacked_params_mapping: