Fix typing order (#39467)

* fix type order

* change all Union[str, dict] to Union[dict, str]

* add hf_parser test && fix test order

* add deepspeed dependency

* replace deepspeed with accelerator

src/transformers/modeling_utils.py

@@ -1395,7 +1395,7 @@ def _get_torch_dtype(
 
 def _get_device_map(
     model: "PreTrainedModel",
-    device_map: Optional[Union[str, dict]],
+    device_map: Optional[Union[dict, str]],
     max_memory: Optional[dict],
     hf_quantizer: Optional[HfQuantizer],
     torch_dtype: Optional[torch.dtype],
@@ -2273,7 +2273,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, PushToHubMixin, PeftAdapterMi
         return model
 
     @classmethod
-    def _check_attn_implementation(cls, attn_implementation: Union[str, dict]) -> Union[str, dict]:
+    def _check_attn_implementation(cls, attn_implementation: Union[dict, str]) -> Union[dict, str]:
         """
         Checks that the requested attention implementation exists and tries to get the kernel from hub
         if `attn_implementation` matches hf kernels pattern.
@@ -2321,7 +2321,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, PushToHubMixin, PeftAdapterMi
 
         return attn_implementation
 
-    def set_attention_implementation(self, attn_implementation: Union[str, dict]):
+    def set_attention_implementation(self, attn_implementation: Union[dict, str]):
         """
         Checks and dispatches to the requested attention implementation.
         """

src/transformers/models/modernbert/modeling_modernbert.py

@@ -611,7 +611,7 @@ class ModernBertPreTrainedModel(PreTrainedModel):
             if module.bias is not None:
                 module.bias.data.zero_()
 
-    def set_attention_implementation(self, attn_implementation: Union[str, dict]):
+    def set_attention_implementation(self, attn_implementation: Union[dict, str]):
         """
         Checks and dispatches to hhe requested attention implementation.
         """

src/transformers/models/modernbert/modular_modernbert.py

@@ -811,7 +811,7 @@ class ModernBertPreTrainedModel(PreTrainedModel):
             if module.bias is not None:
                 module.bias.data.zero_()
 
-    def set_attention_implementation(self, attn_implementation: Union[str, dict]):
+    def set_attention_implementation(self, attn_implementation: Union[dict, str]):
         """
         Checks and dispatches to hhe requested attention implementation.
         """

tests/utils/test_hf_argparser.py

@@ -23,6 +23,7 @@ from dataclasses import dataclass, field
 from enum import Enum
 from pathlib import Path
 from typing import Literal, Optional, Union, get_args, get_origin
+from unittest.mock import patch
 
 import yaml
 
@@ -160,7 +161,7 @@ class HfArgumentParserTest(unittest.TestCase):
 
             self.assertEqual(xx, yy)
 
-    def test_basic(self):
+    def test_00_basic(self):
         parser = HfArgumentParser(BasicExample)
 
         expected = argparse.ArgumentParser()
@@ -174,7 +175,7 @@ class HfArgumentParserTest(unittest.TestCase):
         (example,) = parser.parse_args_into_dataclasses(args, look_for_args_file=False)
         self.assertFalse(example.flag)
 
-    def test_with_default(self):
+    def test_01_with_default(self):
         parser = HfArgumentParser(WithDefaultExample)
 
         expected = argparse.ArgumentParser()
@@ -182,7 +183,7 @@ class HfArgumentParserTest(unittest.TestCase):
         expected.add_argument("--baz", default="toto", type=str, help="help message")
         self.argparsersEqual(parser, expected)
 
-    def test_with_default_bool(self):
+    def test_02_with_default_bool(self):
         expected = argparse.ArgumentParser()
         expected.add_argument("--foo", type=string_to_bool, default=False, const=True, nargs="?")
         expected.add_argument("--baz", type=string_to_bool, default=True, const=True, nargs="?")
@@ -217,7 +218,7 @@ class HfArgumentParserTest(unittest.TestCase):
             args = parser.parse_args(["--foo", "False", "--baz", "False", "--opt", "False"])
             self.assertEqual(args, Namespace(foo=False, baz=False, opt=False))
 
-    def test_with_enum(self):
+    def test_03_with_enum(self):
         parser = HfArgumentParser(MixedTypeEnumExample)
 
         expected = argparse.ArgumentParser()
@@ -244,7 +245,7 @@ class HfArgumentParserTest(unittest.TestCase):
         enum_ex = parser.parse_args_into_dataclasses(["--foo", "42"])[0]
         self.assertEqual(enum_ex.foo, MixedTypeEnum.fourtytwo)
 
-    def test_with_literal(self):
+    def test_04_with_literal(self):
         @dataclass
         class LiteralExample:
             foo: Literal["titi", "toto", 42] = "toto"
@@ -269,7 +270,7 @@ class HfArgumentParserTest(unittest.TestCase):
         args = parser.parse_args(["--foo", "42"])
         self.assertEqual(args.foo, 42)
 
-    def test_with_list(self):
+    def test_05_with_list(self):
         parser = HfArgumentParser(ListExample)
 
         expected = argparse.ArgumentParser()
@@ -292,7 +293,7 @@ class HfArgumentParserTest(unittest.TestCase):
         args = parser.parse_args("--foo-int 1 --bar-int 2 3 --foo-str a b c --foo-float 0.1 0.7".split())
         self.assertEqual(args, Namespace(foo_int=[1], bar_int=[2, 3], foo_str=["a", "b", "c"], foo_float=[0.1, 0.7]))
 
-    def test_with_optional(self):
+    def test_06_with_optional(self):
         expected = argparse.ArgumentParser()
         expected.add_argument("--foo", default=None, type=int)
         expected.add_argument("--bar", default=None, type=float, help="help message")
@@ -315,7 +316,7 @@ class HfArgumentParserTest(unittest.TestCase):
             args = parser.parse_args("--foo 12 --bar 3.14 --baz 42 --ces a b c --des 1 2 3".split())
             self.assertEqual(args, Namespace(foo=12, bar=3.14, baz="42", ces=["a", "b", "c"], des=[1, 2, 3]))
 
-    def test_with_required(self):
+    def test_07_with_required(self):
         parser = HfArgumentParser(RequiredExample)
 
         expected = argparse.ArgumentParser()
@@ -330,7 +331,7 @@ class HfArgumentParserTest(unittest.TestCase):
         )
         self.argparsersEqual(parser, expected)
 
-    def test_with_string_literal_annotation(self):
+    def test_08_with_string_literal_annotation(self):
         parser = HfArgumentParser(StringLiteralAnnotationExample)
 
         expected = argparse.ArgumentParser()
@@ -347,7 +348,7 @@ class HfArgumentParserTest(unittest.TestCase):
         expected.add_argument("--foo_str", "--foo-str", nargs="+", default=["Hallo", "Bonjour", "Hello"], type=str)
         self.argparsersEqual(parser, expected)
 
-    def test_parse_dict(self):
+    def test_09_parse_dict(self):
         parser = HfArgumentParser(BasicExample)
 
         args_dict = {
@@ -361,7 +362,7 @@ class HfArgumentParserTest(unittest.TestCase):
         args = BasicExample(**args_dict)
         self.assertEqual(parsed_args, args)
 
-    def test_parse_dict_extra_key(self):
+    def test_10_parse_dict_extra_key(self):
         parser = HfArgumentParser(BasicExample)
 
         args_dict = {
@@ -374,7 +375,7 @@ class HfArgumentParserTest(unittest.TestCase):
 
         self.assertRaises(ValueError, parser.parse_dict, args_dict, allow_extra_keys=False)
 
-    def test_parse_json(self):
+    def test_11_parse_json(self):
         parser = HfArgumentParser(BasicExample)
 
         args_dict_for_json = {
@@ -393,7 +394,7 @@ class HfArgumentParserTest(unittest.TestCase):
         args = BasicExample(**args_dict_for_json)
         self.assertEqual(parsed_args, args)
 
-    def test_parse_yaml(self):
+    def test_12_parse_yaml(self):
         parser = HfArgumentParser(BasicExample)
 
         args_dict_for_yaml = {
@@ -411,12 +412,7 @@ class HfArgumentParserTest(unittest.TestCase):
         args = BasicExample(**args_dict_for_yaml)
         self.assertEqual(parsed_args, args)
 
-    def test_z_integration_training_args(self):
+    def test_13_valid_dict_annotation(self):
-        # make sure that this test executes last in the test suite
-        parser = HfArgumentParser(TrainingArguments)
-        self.assertIsNotNone(parser)
-
-    def test_valid_dict_annotation(self):
         """
         Tests to make sure that `dict` based annotations
         are correctly made in the `TrainingArguments`.
@@ -475,7 +471,7 @@ class HfArgumentParserTest(unittest.TestCase):
             )
 
     @require_torch
-    def test_valid_dict_input_parsing(self):
+    def test_14_valid_dict_input_parsing(self):
         with tempfile.TemporaryDirectory() as tmp_dir:
             args = TrainingArguments(
                 output_dir=tmp_dir,
@@ -483,3 +479,14 @@ class HfArgumentParserTest(unittest.TestCase):
             )
             self.assertEqual(args.accelerator_config.split_batches, True)
             self.assertEqual(args.accelerator_config.gradient_accumulation_kwargs["num_steps"], 2)
+
+    def test_15_integration_training_args(self):
+        parser = HfArgumentParser(TrainingArguments)
+        self.assertIsNotNone(parser)
+
+    @require_torch
+    @patch("sys.argv", ["test.py", "--accelerator_config", '{"gradient_accumulation_kwargs": {"num_steps": 2}}'])
+    def test_16_cli_input_parsing(self):
+        parser = HfArgumentParser(TrainingArguments)
+        training_args = parser.parse_args_into_dataclasses()[0]
+        self.assertEqual(training_args.accelerator_config.gradient_accumulation_kwargs["num_steps"], 2)