[Misc] Add uninitialized params tracking for AutoWeightsLoader (#10327)

Signed-off-by: Isotr0py <2037008807@qq.com>

vllm/model_executor/models/falcon.py

@@ -18,7 +18,7 @@
 """PyTorch Falcon model."""
 
 import math
-from typing import Iterable, List, Optional, Tuple, Union
+from typing import Iterable, List, Optional, Set, Tuple, Union
 
 import torch
 from torch import nn
@@ -473,7 +473,8 @@ class FalconForCausalLM(nn.Module, SupportsPP):
         next_tokens = self.sampler(logits, sampling_metadata)
         return next_tokens
 
-    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
+    def load_weights(self, weights: Iterable[Tuple[str,
+                                                   torch.Tensor]]) -> Set[str]:
         total_num_heads = self.config.num_attention_heads
         if self.config.new_decoder_architecture:
             total_num_kv_heads = self.config.num_kv_heads
@@ -483,6 +484,7 @@ class FalconForCausalLM(nn.Module, SupportsPP):
             total_num_kv_heads = total_num_heads
         num_query_heads_per_kv_head = total_num_heads // total_num_kv_heads
         params_dict = dict(self.named_parameters(remove_duplicate=False))
+        loaded_params: Set[str] = set()
         for name, loaded_weight in weights:
             if name == "lm_head.weight" and self.tie_word_embeddings:
                 # Falcon uses tied embeddings except Falcon-11b.
@@ -519,3 +521,5 @@ class FalconForCausalLM(nn.Module, SupportsPP):
             weight_loader = getattr(param, "weight_loader",
                                     default_weight_loader)
             weight_loader(param, loaded_weight)
+            loaded_params.add(name)
+        return loaded_params