[Serialization] Add is_main_process argument to save_torch_state_dict()

src/huggingface_hub/serialization/_torch.py

@@ -41,6 +41,7 @@ def save_torch_model(
     max_shard_size: Union[int, str] = MAX_SHARD_SIZE,
     metadata: Optional[Dict[str, str]] = None,
     safe_serialization: bool = True,
+    is_main_process: bool = True,
 ):
     """
     Saves a given torch model to disk, handling sharding and shared tensors issues.
@@ -88,6 +89,10 @@ def save_torch_model(
             Whether to save as safetensors, which is the default behavior. If `False`, the shards are saved as pickle.
             Safe serialization is recommended for security reasons. Saving as pickle is deprecated and will be removed
             in a future version.
+        is_main_process (`bool`, *optional*):
+            Whether the process calling this is the main process or not. Useful when in distributed training like
+            TPUs and need to call this function on all processes. In this case, set `is_main_process=True` only on
+            the main process to avoid race conditions. Defaults to True.
 
     Example:
 
@@ -112,6 +117,7 @@ def save_torch_model(
         metadata=metadata,
         safe_serialization=safe_serialization,
         save_directory=save_directory,
+        is_main_process=is_main_process,
     )
 
 
@@ -124,6 +130,7 @@ def save_torch_state_dict(
     max_shard_size: Union[int, str] = MAX_SHARD_SIZE,
     metadata: Optional[Dict[str, str]] = None,
     safe_serialization: bool = True,
+    is_main_process: bool = True,
 ) -> None:
     """
     Save a model state dictionary to the disk, handling sharding and shared tensors issues.
@@ -171,7 +178,10 @@ def save_torch_state_dict(
             Whether to save as safetensors, which is the default behavior. If `False`, the shards are saved as pickle.
             Safe serialization is recommended for security reasons. Saving as pickle is deprecated and will be removed
             in a future version.
-
+        is_main_process (`bool`, *optional*):
+            Whether the process calling this is the main process or not. Useful when in distributed training like
+            TPUs and need to call this function on all processes. In this case, set `is_main_process=True` only on
+            the main process to avoid race conditions. Defaults to True.
     Example:
 
     ```py
@@ -222,15 +232,18 @@ def save_torch_state_dict(
         state_dict, filename_pattern=filename_pattern, max_shard_size=max_shard_size
     )
 
-    # Clean the folder from previous save
-    existing_files_regex = re.compile(filename_pattern.format(suffix=r"(-\d{5}-of-\d{5})?") + r"(\.index\.json)?")
-    for filename in os.listdir(save_directory):
-        if existing_files_regex.match(filename):
-            try:
-                logger.debug(f"Removing existing file '{filename}' from folder.")
-                os.remove(os.path.join(save_directory, filename))
-            except Exception as e:
-                logger.warning(f"Error when trying to remove existing '{filename}' from folder: {e}. Continuing...")
+    # Only main process should clean up existing files to avoid race conditions in distributed environment
+    if is_main_process:
+        existing_files_regex = re.compile(filename_pattern.format(suffix=r"(-\d{5}-of-\d{5})?") + r"(\.index\.json)?")
+        for filename in os.listdir(save_directory):
+            if existing_files_regex.match(filename):
+                try:
+                    logger.debug(f"Removing existing file '{filename}' from folder.")
+                    os.remove(os.path.join(save_directory, filename))
+                except Exception as e:
+                    logger.warning(
+                        f"Error when trying to remove existing '{filename}' from folder: {e}. Continuing..."
+                    )
 
     # Save each shard
     per_file_metadata = {"format": "pt"}
@@ -442,7 +455,7 @@ def storage_ptr(tensor: "torch.Tensor") -> Union[int, Tuple[Any, ...]]:
         from torch.utils._python_dispatch import is_traceable_wrapper_subclass
 
         if is_traceable_wrapper_subclass(tensor):
-            return _get_unique_id(tensor)
+            return _get_unique_id(tensor)  # type: ignore
     except ImportError:
         # for torch version less than 2.1, we can fallback to original implementation
         pass

tests/test_serialization.py

@@ -264,6 +264,7 @@ def test_save_torch_model(mocker: MockerFixture, tmp_path: Path) -> None:
         max_shard_size="3GB",
         metadata={"foo": "bar"},
         safe_serialization=True,
+        is_main_process=True,
     )
     safe_state_dict_mock.assert_called_once_with(
         state_dict=model_mock.state_dict.return_value,
@@ -273,6 +274,7 @@ def test_save_torch_model(mocker: MockerFixture, tmp_path: Path) -> None:
         max_shard_size="3GB",
         metadata={"foo": "bar"},
         safe_serialization=True,
+        is_main_process=True,
     )
 
 
@@ -472,3 +474,27 @@ def test_save_torch_state_dict_delete_existing_files(
     assert (tmp_path / "pytorch_model-00001-of-00003.bin").is_file()
     assert (tmp_path / "pytorch_model-00002-of-00003.bin").is_file()
     assert (tmp_path / "pytorch_model-00003-of-00003.bin").is_file()
+
+
+def test_save_torch_state_dict_not_main_process(
+    tmp_path: Path,
+    torch_state_dict: Dict[str, "torch.Tensor"],
+) -> None:
+    """
+    Test that previous files in the directory are not deleted when is_main_process=False.
+    When is_main_process=True, previous files should be deleted,
+    this is already tested in `test_save_torch_state_dict_delete_existing_files`.
+    """
+    # Create some .safetensors files before saving a new state dict.
+    (tmp_path / "model.safetensors").touch()
+    (tmp_path / "model-00001-of-00002.safetensors").touch()
+    (tmp_path / "model-00002-of-00002.safetensors").touch()
+    (tmp_path / "model.safetensors.index.json").touch()
+    # Save with is_main_process=False
+    save_torch_state_dict(torch_state_dict, tmp_path, is_main_process=False)
+
+    # Previous files should still exist (not deleted)
+    assert (tmp_path / "model.safetensors").is_file()
+    assert (tmp_path / "model-00001-of-00002.safetensors").is_file()
+    assert (tmp_path / "model-00002-of-00002.safetensors").is_file()
+    assert (tmp_path / "model.safetensors.index.json").is_file()