Release: V4.5.1

* added model_kwargs to infer_framework_from_model

* added model_kwargs to tokenizer

* added use_auth_token as named parameter

* added dynamic get for use_auth_token

setup.py

@@ -305,7 +305,7 @@ install_requires = [
 
 setup(
     name="transformers",
-    version="4.5.0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    version="4.5.1",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
     author="Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Sam Shleifer, Patrick von Platen, Sylvain Gugger, Google AI Language Team Authors, Open AI team Authors, Facebook AI Authors, Carnegie Mellon University Authors",
     author_email="thomas@huggingface.co",
     description="State-of-the-art Natural Language Processing for TensorFlow 2.0 and PyTorch",

src/transformers/__init__.py

@@ -22,7 +22,7 @@
 # to defer the actual importing for when the objects are requested. This way `import transformers` provides the names
 # in the namespace without actually importing anything (and especially none of the backends).
 
-__version__ = "4.5.0"
+__version__ = "4.5.1"
 
 # Work around to update TensorFlow's absl.logging threshold which alters the
 # default Python logging output behavior when present.

src/transformers/configuration_utils.py

@@ -399,10 +399,11 @@ class PretrainedConfig(object):
 
         """
         config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
-        if config_dict.get("model_type", False) and hasattr(cls, "model_type"):
-            assert (
-                config_dict["model_type"] == cls.model_type
-            ), f"You tried to initiate a model of type '{cls.model_type}' with a pretrained model of type '{config_dict['model_type']}'"
+        if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
+            logger.warn(
+                f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
+                f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
+            )
 
         return cls.from_dict(config_dict, **kwargs)
 

src/transformers/pipelines/__init__.py

@@ -246,6 +246,7 @@ def pipeline(
     framework: Optional[str] = None,
     revision: Optional[str] = None,
     use_fast: bool = True,
+    use_auth_token: Optional[Union[str, bool]] = None,
     model_kwargs: Dict[str, Any] = {},
     **kwargs
 ) -> Pipeline:
@@ -308,6 +309,10 @@ def pipeline(
             artifacts on huggingface.co, so ``revision`` can be any identifier allowed by git.
         use_fast (:obj:`bool`, `optional`, defaults to :obj:`True`):
             Whether or not to use a Fast tokenizer if possible (a :class:`~transformers.PreTrainedTokenizerFast`).
+        use_auth_token (:obj:`str` or `bool`, `optional`):
+            The token to use as HTTP bearer authorization for remote files. If :obj:`True`, will use the token
+            generated when running :obj:`transformers-cli login` (stored in :obj:`~/.huggingface`).
+            revision(:obj:`str`, `optional`, defaults to :obj:`"main"`):
         model_kwargs:
             Additional dictionary of keyword arguments passed along to the model's :obj:`from_pretrained(...,
             **model_kwargs)` function.
@@ -367,6 +372,9 @@ def pipeline(
 
     task_class, model_class = targeted_task["impl"], targeted_task[framework]
 
+    # Retrieve use_auth_token and add it to model_kwargs to be used in .from_pretrained
+    model_kwargs["use_auth_token"] = model_kwargs.get("use_auth_token", use_auth_token)
+
     # Instantiate tokenizer if needed
     if isinstance(tokenizer, (str, tuple)):
         if isinstance(tokenizer, tuple):
@@ -377,12 +385,12 @@ def pipeline(
             )
         else:
             tokenizer = AutoTokenizer.from_pretrained(
-                tokenizer, revision=revision, use_fast=use_fast, _from_pipeline=task
+                tokenizer, revision=revision, use_fast=use_fast, _from_pipeline=task, **model_kwargs
             )
 
     # Instantiate config if needed
     if isinstance(config, str):
-        config = AutoConfig.from_pretrained(config, revision=revision, _from_pipeline=task)
+        config = AutoConfig.from_pretrained(config, revision=revision, _from_pipeline=task, **model_kwargs)
 
     # Instantiate modelcard if needed
     if isinstance(modelcard, str):

src/transformers/pipelines/base.py

@@ -48,7 +48,7 @@ logger = logging.get_logger(__name__)
 
 
 def infer_framework_from_model(
-    model, model_classes: Optional[Dict[str, type]] = None, revision: Optional[str] = None, task: Optional[str] = None
+    model, model_classes: Optional[Dict[str, type]] = None, task: Optional[str] = None, **model_kwargs
 ):
     """
     Select framework (TensorFlow or PyTorch) to use from the :obj:`model` passed. Returns a tuple (framework, model).
@@ -65,10 +65,11 @@ def infer_framework_from_model(
             from.
         model_classes (dictionary :obj:`str` to :obj:`type`, `optional`):
             A mapping framework to class.
-        revision (:obj:`str`, `optional`):
-            The specific model version to use. It can be a branch name, a tag name, or a commit id, since we use a
-            git-based system for storing models and other artifacts on huggingface.co, so ``revision`` can be any
-            identifier allowed by git.
+        task (:obj:`str`):
+            The task defining which pipeline will be returned.
+        model_kwargs:
+            Additional dictionary of keyword arguments passed along to the model's :obj:`from_pretrained(...,
+            **model_kwargs)` function.
 
     Returns:
         :obj:`Tuple`: A tuple framework, model.
@@ -80,19 +81,20 @@ def infer_framework_from_model(
             "To install PyTorch, read the instructions at https://pytorch.org/."
         )
     if isinstance(model, str):
+        model_kwargs["_from_pipeline"] = task
         if is_torch_available() and not is_tf_available():
             model_class = model_classes.get("pt", AutoModel)
-            model = model_class.from_pretrained(model, revision=revision, _from_pipeline=task)
+            model = model_class.from_pretrained(model, **model_kwargs)
         elif is_tf_available() and not is_torch_available():
             model_class = model_classes.get("tf", TFAutoModel)
-            model = model_class.from_pretrained(model, revision=revision, _from_pipeline=task)
+            model = model_class.from_pretrained(model, **model_kwargs)
         else:
             try:
                 model_class = model_classes.get("pt", AutoModel)
-                model = model_class.from_pretrained(model, revision=revision, _from_pipeline=task)
+                model = model_class.from_pretrained(model, **model_kwargs)
             except OSError:
                 model_class = model_classes.get("tf", TFAutoModel)
-                model = model_class.from_pretrained(model, revision=revision, _from_pipeline=task)
+                model = model_class.from_pretrained(model, **model_kwargs)
 
     framework = "tf" if model.__class__.__name__.startswith("TF") else "pt"
     return framework, model

tests/test_modeling_bert_generation.py

@@ -231,13 +231,7 @@ class BertGenerationEncoderTester:
         self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
 
     def prepare_config_and_inputs_for_common(self):
-        config_and_inputs = self.prepare_config_and_inputs()
-        (
-            config,
-            input_ids,
-            input_mask,
-            token_labels,
-        ) = config_and_inputs
+        config, input_ids, input_mask, token_labels = self.prepare_config_and_inputs()
         inputs_dict = {"input_ids": input_ids, "attention_mask": input_mask}
         return config, inputs_dict
 
@@ -259,6 +253,11 @@ class BertGenerationEncoderTest(ModelTesterMixin, GenerationTesterMixin, unittes
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_model(*config_and_inputs)
 
+    def test_model_as_bert(self):
+        config, input_ids, input_mask, token_labels = self.model_tester.prepare_config_and_inputs()
+        config.model_type = "bert"
+        self.model_tester.create_and_check_model(config, input_ids, input_mask, token_labels)
+
     def test_model_as_decoder(self):
         config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
         self.model_tester.create_and_check_model_as_decoder(*config_and_inputs)

tests/test_modeling_common.py

@@ -22,9 +22,9 @@ import tempfile
 import unittest
 from typing import List, Tuple
 
-from transformers import is_torch_available
+from transformers import is_torch_available, logging
 from transformers.file_utils import WEIGHTS_NAME
-from transformers.testing_utils import require_torch, require_torch_multi_gpu, slow, torch_device
+from transformers.testing_utils import CaptureLogger, require_torch, require_torch_multi_gpu, slow, torch_device
 
 
 if is_torch_available():
@@ -1295,6 +1295,7 @@ class ModelUtilsTest(unittest.TestCase):
         model = T5ForConditionalGeneration.from_pretrained(TINY_T5)
         self.assertIsNotNone(model)
 
-        with self.assertRaises(Exception) as context:
+        logger = logging.get_logger("transformers.configuration_utils")
+        with CaptureLogger(logger) as cl:
             BertModel.from_pretrained(TINY_T5)
-        self.assertTrue("You tried to initiate a model of type" in str(context.exception))
+        self.assertTrue("You are using a model of type t5 to instantiate a model of type bert" in cl.out)