diff --git a/pytorch_transformers/modeling_utils.py b/pytorch_transformers/modeling_utils.py index a4e1a44c..0a4bfa7b 100644 --- a/pytorch_transformers/modeling_utils.py +++ b/pytorch_transformers/modeling_utils.py @@ -373,7 +373,8 @@ class PreTrainedModel(nn.Module): if config is None: config, model_kwargs = cls.config_class.from_pretrained( pretrained_model_name_or_path, *model_args, - return_unused_args=True, **kwargs + cache_dir=cache_dir, return_unused_args=True, + **kwargs ) else: model_kwargs = kwargs