From 3e1a279a73b4884145b4bf830f3bfebbfa8fb027 Mon Sep 17 00:00:00 2001 From: Yuanyuan Chen Date: Thu, 18 Sep 2025 19:02:44 +0800 Subject: [PATCH 1/2] Fix outdated version checks of accelerator Signed-off-by: Yuanyuan Chen --- src/transformers/trainer.py | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-) diff --git a/src/transformers/trainer.py b/src/transformers/trainer.py index 97de1d6d5397..10fa8cad13cb 100755 --- a/src/transformers/trainer.py +++ b/src/transformers/trainer.py @@ -241,10 +241,9 @@ DATA_SAMPLERS = [RandomSampler] if version.parse(accelerate_version) > version.parse("1.3.0"): from accelerate.utils import TorchTensorParallelPlugin - if version.parse(accelerate_version) > version.parse("0.23.0"): - from accelerate.data_loader import SeedableRandomSampler + from accelerate.data_loader import SeedableRandomSampler - DATA_SAMPLERS += [SeedableRandomSampler] + DATA_SAMPLERS += [SeedableRandomSampler] if is_deepspeed_available(): from accelerate.utils import DeepSpeedSchedulerWrapper @@ -4196,9 +4195,7 @@ def save_model(self, output_dir: Optional[str] = None, _internal_call: bool = Fa elif (tp_size := getattr(self.model, "_tp_size", 0)) is not None and tp_size > 1: self._save(output_dir) elif self.is_fsdp_enabled: - if ("FULL_STATE_DICT" in str(self.accelerator.state.fsdp_plugin.state_dict_type)) and ( - version.parse(accelerate_version) > version.parse("0.24.1") - ): + if "FULL_STATE_DICT" in str(self.accelerator.state.fsdp_plugin.state_dict_type): state_dict = self.accelerator.get_state_dict(self.model) if self.args.should_save: self._save(output_dir, state_dict=state_dict) From a49e6f5fe3f0952226beaec7e2b75ec2c4a5dc86 Mon Sep 17 00:00:00 2001 From: Yuanyuan Chen Date: Thu, 18 Sep 2025 19:38:05 +0800 Subject: [PATCH 2/2] Fix outdated version checks of accelerator Signed-off-by: Yuanyuan Chen --- tests/fsdp/test_fsdp.py | 14 +------------- 1 file changed, 1 insertion(+), 13 deletions(-) diff --git a/tests/fsdp/test_fsdp.py b/tests/fsdp/test_fsdp.py index a932a1fbac67..6a4060b0a731 100644 --- a/tests/fsdp/test_fsdp.py +++ b/tests/fsdp/test_fsdp.py @@ -88,22 +88,11 @@ def get_master_port(real_launcher=False): if is_torch_available(): - from tests.trainer.test_trainer import ( # noqa - RegressionModelConfig, - RegressionPreTrainedModel, - ) - # hack to restore original logging level pre #21700 get_regression_trainer = partial(tests.trainer.test_trainer.get_regression_trainer, log_level="info") -require_fsdp_version = require_fsdp if is_accelerate_available(): - from accelerate.utils.constants import ( - FSDP_PYTORCH_VERSION, - FSDP_SHARDING_STRATEGY, - ) - - require_fsdp_version = partial(require_fsdp, min_version=FSDP_PYTORCH_VERSION) + from accelerate.utils.constants import FSDP_SHARDING_STRATEGY FSDP2_ACCELERATE_VERSION = "1.6.0" @@ -142,7 +131,6 @@ def _parameterized_custom_name_func(func, param_num, param): @require_accelerate @require_torch_accelerator -@require_fsdp_version class TrainerIntegrationFSDP(TestCasePlus, TrainerIntegrationCommon): def setUp(self): super().setUp()