Merge branch 'main' into base-model-loading

src/transformers/modeling_utils.py

@@ -819,15 +819,15 @@ def _load_state_dict_into_meta_model(
     is_quantized = hf_quantizer is not None
 
     is_torch_e4m3fn_available = hasattr(torch, "float8_e4m3fn")
-
     for serialized_param_name, empty_param in state_dict.items():
+        # we shouldn't rename the key (add/remove prefix) before checking its value in expected keys but it's fine for legacy params
+        serialized_param_name, _ = model._fix_state_dict_key_on_load(serialized_param_name)
+        if serialized_param_name not in expected_keys:
+            continue
         # serialized_param_name is the raw, serialized name
         # fixed_param_name is the model's equivalent
         fixed_param_name, _ = model.rename_key(serialized_param_name)
 
-        if fixed_param_name not in expected_keys:
-            continue
-
         # we need to use serialized_param_name as file pointer is untouched
         param = (
             file_pointer.get_slice(serialized_param_name)