Fix _load_state_dict_into_meta_model with device_map=None (huggingfac…

…e#36488) * Fix _load_state_dict_into_meta_model with device_map=None * Update src/transformers/modeling_utils.py
garrett361 · Mar 4, 2025 · b1c0145 · b1c0145
1 parent 4f8c635
commit b1c0145
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/src/transformers/modeling_utils.py b/src/transformers/modeling_utils.py
@@ -785,8 +785,8 @@ def _load_state_dict_into_meta_model(
     tensor_device = None
     if device_map is not None and device_map.get("", None) is not None:
         tensor_device = device_map[""].index if isinstance(device_map[""], torch.device) else device_map[""]
-
-    device_map_regex = "|".join(sorted(device_map.keys(), reverse=True))
+    if device_map is not None:
+        device_map_regex = "|".join(sorted(device_map.keys(), reverse=True))
 
     # we need this later to initialize tensor parallelism
     if device_mesh is not None: