Tokenizers: ability to load from model subfolder (#8586)

* <small>tiny typo</small>

* Tokenizers: ability to load from model subfolder

* use subfolder for local files as well

* Uniformize model shortcut name => model id

* from s3 => from huggingface.co

Co-authored-by: Quentin Lhoest <lhoest.q@gmail.com>

docs/source/pretrained_models.rst

@@ -3,11 +3,11 @@ Pretrained models
 
 Here is the full list of the currently provided pretrained models together with a short presentation of each model.
 
-For a list that includes community-uploaded models, refer to `https://huggingface.co/models
+For a list that includes all community-uploaded models, refer to `https://huggingface.co/models
 <https://huggingface.co/models>`__.
 
 +--------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
-| Architecture       | Shortcut name                                              | Details of the model                                                                                                                  |
+| Architecture       | Model id                                                   | Details of the model                                                                                                                  |
 +====================+============================================================+=======================================================================================================================================+
 | BERT               | ``bert-base-uncased``                                      | | 12-layer, 768-hidden, 12-heads, 110M parameters.                                                                                    |
 |                    |                                                            | | Trained on lower-cased English text.                                                                                                |

examples/adversarial/run_hans.py

@@ -57,7 +57,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/bert-loses-patience/run_glue_with_pabee.py

@@ -476,7 +476,7 @@ def main():
         "--cache_dir",
         default="",
         type=str,
-        help="Where do you want to store the pre-trained models downloaded from s3",
+        help="Where do you want to store the pre-trained models downloaded from huggingface.co",
     )
     parser.add_argument(
         "--max_seq_length",

examples/bertology/run_bertology.py

@@ -298,7 +298,7 @@ def main():
         "--cache_dir",
         default=None,
         type=str,
-        help="Where do you want to store the pre-trained models downloaded from s3",
+        help="Where do you want to store the pre-trained models downloaded from huggingface.co",
     )
     parser.add_argument(
         "--data_subset", type=int, default=-1, help="If > 0: limit the data to a subset of data_subset instances."

examples/contrib/legacy/run_language_modeling.py

@@ -81,7 +81,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/contrib/mm-imdb/run_mmimdb.py

@@ -350,7 +350,7 @@ def main():
         "--cache_dir",
         default=None,
         type=str,
-        help="Where do you want to store the pre-trained models downloaded from s3",
+        help="Where do you want to store the pre-trained models downloaded from huggingface.co",
     )
     parser.add_argument(
         "--max_seq_length",

examples/deebert/run_glue_deebert.py

@@ -452,7 +452,7 @@ def main():
         "--cache_dir",
         default="",
         type=str,
-        help="Where do you want to store the pre-trained models downloaded from s3",
+        help="Where do you want to store the pre-trained models downloaded from huggingface.co",
     )
     parser.add_argument(
         "--max_seq_length",

examples/distillation/run_squad_w_distillation.py

@@ -578,7 +578,7 @@ def main():
         "--cache_dir",
         default="",
         type=str,
-        help="Where do you want to store the pre-trained models downloaded from s3",
+        help="Where do you want to store the pre-trained models downloaded from huggingface.co",
     )
 
     parser.add_argument(

examples/language-modeling/run_clm.py

@@ -76,7 +76,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
     use_fast_tokenizer: bool = field(
         default=True,

examples/language-modeling/run_mlm.py

@@ -74,7 +74,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
     use_fast_tokenizer: bool = field(
         default=True,

examples/language-modeling/run_mlm_wwm.py

@@ -76,7 +76,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
     use_fast_tokenizer: bool = field(
         default=True,

examples/language-modeling/run_plm.py

@@ -64,7 +64,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
     use_fast_tokenizer: bool = field(
         default=True,

examples/lightning_base.py

@@ -236,7 +236,7 @@ class BaseTransformer(pl.LightningModule):
             "--cache_dir",
             default="",
             type=str,
-            help="Where do you want to store the pre-trained models downloaded from s3",
+            help="Where do you want to store the pre-trained models downloaded from huggingface.co",
         )
         parser.add_argument(
             "--encoder_layerdrop",

examples/movement-pruning/masked_run_glue.py

@@ -620,7 +620,7 @@ def main():
         "--cache_dir",
         default="",
         type=str,
-        help="Where do you want to store the pre-trained models downloaded from s3",
+        help="Where do you want to store the pre-trained models downloaded from huggingface.co",
     )
     parser.add_argument(
         "--max_seq_length",

examples/movement-pruning/masked_run_squad.py

@@ -725,7 +725,7 @@ def main():
         "--cache_dir",
         default="",
         type=str,
-        help="Where do you want to store the pre-trained models downloaded from s3",
+        help="Where do you want to store the pre-trained models downloaded from huggingface.co",
     )
 
     parser.add_argument(

examples/multiple-choice/run_multiple_choice.py

@@ -61,7 +61,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/multiple-choice/run_tf_multiple_choice.py

@@ -65,7 +65,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/question-answering/run_squad.py

@@ -532,7 +532,7 @@ def main():
         "--cache_dir",
         default="",
         type=str,
-        help="Where do you want to store the pre-trained models downloaded from s3",
+        help="Where do you want to store the pre-trained models downloaded from huggingface.co",
     )
 
     parser.add_argument(

examples/question-answering/run_squad_trainer.py

@@ -51,7 +51,8 @@ class ModelArguments:
     # If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
     # or just modify its tokenizer_config.json.
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/question-answering/run_tf_squad.py

@@ -63,7 +63,8 @@ class ModelArguments:
     # If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
     # or just modify its tokenizer_config.json.
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/rag/finetune.sh

@@ -7,7 +7,7 @@ export PYTHONPATH="../":"${PYTHONPATH}"
 python examples/rag/finetune.py \
     --data_dir $DATA_DIR \
     --output_dir $OUTPUT_DIR \
-    --model_name_or_path $MODLE_NAME_OR_PATH \
+    --model_name_or_path $MODEL_NAME_OR_PATH \
     --model_type rag_sequence \
     --fp16 \
     --gpus 8 \

examples/seq2seq/finetune_trainer.py

@@ -43,7 +43,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
     freeze_encoder: bool = field(default=False, metadata={"help": "Whether tp freeze the encoder."})
     freeze_embeds: bool = field(default=False, metadata={"help": "Whether  to freeze the embeddings."})

examples/text-classification/run_glue.py

@@ -124,7 +124,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
     use_fast_tokenizer: bool = field(
         default=True,

examples/text-classification/run_tf_glue.py

@@ -117,7 +117,8 @@ class ModelArguments:
     # If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
     # or just modify its tokenizer_config.json.
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/text-classification/run_tf_text_classification.py

@@ -182,7 +182,8 @@ class ModelArguments:
     # If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
     # or just modify its tokenizer_config.json.
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/text-classification/run_xnli.py

@@ -406,7 +406,7 @@ def main():
         "--cache_dir",
         default=None,
         type=str,
-        help="Where do you want to store the pre-trained models downloaded from s3",
+        help="Where do you want to store the pre-trained models downloaded from huggingface.co",
     )
     parser.add_argument(
         "--max_seq_length",

examples/token-classification/run_ner.py

@@ -60,7 +60,8 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/token-classification/run_ner_old.py

@@ -65,7 +65,8 @@ class ModelArguments:
     # If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
     # or just modify its tokenizer_config.json.
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

examples/token-classification/run_tf_ner.py

@@ -67,7 +67,8 @@ class ModelArguments:
     # If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
     # or just modify its tokenizer_config.json.
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
 
 

hubconf.py

@@ -25,7 +25,7 @@ def config(*args, **kwargs):
                 # Using torch.hub !
                 import torch
 
-                config = torch.hub.load('huggingface/transformers', 'config', 'bert-base-uncased')  # Download configuration from S3 and cache.
+                config = torch.hub.load('huggingface/transformers', 'config', 'bert-base-uncased')  # Download configuration from huggingface.co and cache.
                 config = torch.hub.load('huggingface/transformers', 'config', './test/bert_saved_model/')  # E.g. config (or model) was saved using `save_pretrained('./test/saved_model/')`
                 config = torch.hub.load('huggingface/transformers', 'config', './test/bert_saved_model/my_configuration.json')
                 config = torch.hub.load('huggingface/transformers', 'config', 'bert-base-uncased', output_attentions=True, foo=False)
@@ -45,7 +45,7 @@ def tokenizer(*args, **kwargs):
         # Using torch.hub !
         import torch
 
-        tokenizer = torch.hub.load('huggingface/transformers', 'tokenizer', 'bert-base-uncased')    # Download vocabulary from S3 and cache.
+        tokenizer = torch.hub.load('huggingface/transformers', 'tokenizer', 'bert-base-uncased')    # Download vocabulary from huggingface.co and cache.
         tokenizer = torch.hub.load('huggingface/transformers', 'tokenizer', './test/bert_saved_model/')  # E.g. tokenizer was saved using `save_pretrained('./test/saved_model/')`
 
     """
@@ -59,7 +59,7 @@ def model(*args, **kwargs):
             # Using torch.hub !
             import torch
 
-            model = torch.hub.load('huggingface/transformers', 'model', 'bert-base-uncased')    # Download model and configuration from S3 and cache.
+            model = torch.hub.load('huggingface/transformers', 'model', 'bert-base-uncased')    # Download model and configuration from huggingface.co and cache.
             model = torch.hub.load('huggingface/transformers', 'model', './test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
             model = torch.hub.load('huggingface/transformers', 'model', 'bert-base-uncased', output_attentions=True)  # Update configuration during loading
             assert model.config.output_attentions == True
@@ -78,7 +78,7 @@ def modelWithLMHead(*args, **kwargs):
         # Using torch.hub !
         import torch
 
-        model = torch.hub.load('huggingface/transformers', 'modelWithLMHead', 'bert-base-uncased')    # Download model and configuration from S3 and cache.
+        model = torch.hub.load('huggingface/transformers', 'modelWithLMHead', 'bert-base-uncased')    # Download model and configuration from huggingface.co and cache.
         model = torch.hub.load('huggingface/transformers', 'modelWithLMHead', './test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
         model = torch.hub.load('huggingface/transformers', 'modelWithLMHead', 'bert-base-uncased', output_attentions=True)  # Update configuration during loading
         assert model.config.output_attentions == True
@@ -96,7 +96,7 @@ def modelForSequenceClassification(*args, **kwargs):
             # Using torch.hub !
             import torch
 
-            model = torch.hub.load('huggingface/transformers', 'modelForSequenceClassification', 'bert-base-uncased')    # Download model and configuration from S3 and cache.
+            model = torch.hub.load('huggingface/transformers', 'modelForSequenceClassification', 'bert-base-uncased')    # Download model and configuration from huggingface.co and cache.
             model = torch.hub.load('huggingface/transformers', 'modelForSequenceClassification', './test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
             model = torch.hub.load('huggingface/transformers', 'modelForSequenceClassification', 'bert-base-uncased', output_attentions=True)  # Update configuration during loading
             assert model.config.output_attentions == True
@@ -115,7 +115,7 @@ def modelForQuestionAnswering(*args, **kwargs):
         # Using torch.hub !
         import torch
 
-        model = torch.hub.load('huggingface/transformers', 'modelForQuestionAnswering', 'bert-base-uncased')    # Download model and configuration from S3 and cache.
+        model = torch.hub.load('huggingface/transformers', 'modelForQuestionAnswering', 'bert-base-uncased')    # Download model and configuration from huggingface.co and cache.
         model = torch.hub.load('huggingface/transformers', 'modelForQuestionAnswering', './test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
         model = torch.hub.load('huggingface/transformers', 'modelForQuestionAnswering', 'bert-base-uncased', output_attentions=True)  # Update configuration during loading
         assert model.config.output_attentions == True

src/transformers/commands/user.py

@@ -31,7 +31,7 @@ class UserCommands(BaseTransformersCLICommand):
         ls_parser.add_argument("--organization", type=str, help="Optional: organization namespace.")
         ls_parser.set_defaults(func=lambda args: ListObjsCommand(args))
         rm_parser = s3_subparsers.add_parser("rm")
-        rm_parser.add_argument("filename", type=str, help="individual object filename to delete from S3.")
+        rm_parser.add_argument("filename", type=str, help="individual object filename to delete from huggingface.co.")
         rm_parser.add_argument("--organization", type=str, help="Optional: organization namespace.")
         rm_parser.set_defaults(func=lambda args: DeleteObjCommand(args))
         upload_parser = s3_subparsers.add_parser("upload", help="Upload a file to S3.")

src/transformers/configuration_utils.py

@@ -291,10 +291,9 @@ class PretrainedConfig(object):
             pretrained_model_name_or_path (:obj:`str`):
                 This can be either:
 
-                - the `shortcut name` of a pretrained model configuration to load from cache or download, e.g.,
+                - a string, the `model id` of a pretrained model configuration hosted inside a model repo on
-                  ``bert-base-uncased``.
+                  huggingface.co. Valid model ids can be located at the root-level, like ``bert-base-uncased``, or
-                - the `identifier name` of a pretrained model configuration that was uploaded to our S3 by any user,
+                  namespaced under a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                  e.g., ``dbmdz/bert-base-german-cased``.
                 - a path to a `directory` containing a configuration file saved using the
                   :func:`~transformers.PretrainedConfig.save_pretrained` method, e.g., ``./my_model_directory/``.
                 - a path or url to a saved configuration JSON `file`, e.g.,
@@ -333,7 +332,7 @@ class PretrainedConfig(object):
 
             # We can't instantiate directly the base class `PretrainedConfig` so let's show the examples on a
             # derived class: BertConfig
-            config = BertConfig.from_pretrained('bert-base-uncased')    # Download configuration from S3 and cache.
+            config = BertConfig.from_pretrained('bert-base-uncased')    # Download configuration from huggingface.co and cache.
             config = BertConfig.from_pretrained('./test/saved_model/')  # E.g. config (or model) was saved using `save_pretrained('./test/saved_model/')`
             config = BertConfig.from_pretrained('./test/saved_model/my_configuration.json')
             config = BertConfig.from_pretrained('bert-base-uncased', output_attentions=True, foo=False)

src/transformers/file_utils.py

@@ -855,7 +855,9 @@ def is_remote_url(url_or_filename):
     return parsed.scheme in ("http", "https")
 
 
-def hf_bucket_url(model_id: str, filename: str, revision: Optional[str] = None, mirror=None) -> str:
+def hf_bucket_url(
+    model_id: str, filename: str, subfolder: Optional[str] = None, revision: Optional[str] = None, mirror=None
+) -> str:
     """
     Resolve a model identifier, a file name, and an optional revision id, to a huggingface.co-hosted url, redirecting
     to Cloudfront (a Content Delivery Network, or CDN) for large files.
@@ -872,6 +874,9 @@ def hf_bucket_url(model_id: str, filename: str, revision: Optional[str] = None,
     its sha1 if stored in git, or its sha256 if stored in git-lfs. Files cached locally from transformers before v3.5.0
     are not shared with those new files, because the cached file's name contains a hash of the url (which changed).
     """
+    if subfolder is not None:
+        filename = f"{subfolder}/{filename}"
+
     if mirror:
         endpoint = PRESET_MIRROR_DICT.get(mirror, mirror)
         legacy_format = "/" not in model_id

src/transformers/generation_tf_utils.py

@@ -148,12 +148,12 @@ class TFGenerationMixin:
         Examples::
 
             tokenizer = AutoTokenizer.from_pretrained('distilgpt2')   # Initialize tokenizer
-            model = TFAutoModelWithLMHead.from_pretrained('distilgpt2')    # Download model and configuration from S3 and cache.
+            model = TFAutoModelWithLMHead.from_pretrained('distilgpt2')    # Download model and configuration from huggingface.co and cache.
             outputs = model.generate(max_length=40)  # do greedy decoding
             print('Generated: {}'.format(tokenizer.decode(outputs[0], skip_special_tokens=True)))
 
             tokenizer = AutoTokenizer.from_pretrained('openai-gpt')   # Initialize tokenizer
-            model = TFAutoModelWithLMHead.from_pretrained('openai-gpt')    # Download model and configuration from S3 and cache.
+            model = TFAutoModelWithLMHead.from_pretrained('openai-gpt')    # Download model and configuration from huggingface.co and cache.
             input_context = 'The dog'
             input_ids = tokenizer.encode(input_context, return_tensors='tf')  # encode input context
             outputs = model.generate(input_ids=input_ids, num_beams=5, num_return_sequences=3, temperature=1.5)  # generate 3 independent sequences using beam search decoding (5 beams) with sampling from initial context 'The dog'
@@ -161,7 +161,7 @@ class TFGenerationMixin:
                 print('Generated {}: {}'.format(i, tokenizer.decode(outputs[i], skip_special_tokens=True)))
 
             tokenizer = AutoTokenizer.from_pretrained('distilgpt2')   # Initialize tokenizer
-            model = TFAutoModelWithLMHead.from_pretrained('distilgpt2')    # Download model and configuration from S3 and cache.
+            model = TFAutoModelWithLMHead.from_pretrained('distilgpt2')    # Download model and configuration from huggingface.co and cache.
             input_context = 'The dog'
             input_ids = tokenizer.encode(input_context, return_tensors='tf')  # encode input context
             outputs = model.generate(input_ids=input_ids, max_length=40, temperature=0.7, num_return_sequences=3, do_sample=True)  # generate 3 candidates using sampling
@@ -169,14 +169,14 @@ class TFGenerationMixin:
                 print('Generated {}: {}'.format(i, tokenizer.decode(outputs[i], skip_special_tokens=True)))
 
             tokenizer = AutoTokenizer.from_pretrained('ctrl')   # Initialize tokenizer
-            model = TFAutoModelWithLMHead.from_pretrained('ctrl')    # Download model and configuration from S3 and cache.
+            model = TFAutoModelWithLMHead.from_pretrained('ctrl')    # Download model and configuration from huggingface.co and cache.
             input_context = 'Legal My neighbor is'  # "Legal" is one of the control codes for ctrl
             input_ids = tokenizer.encode(input_context, return_tensors='tf')  # encode input context
             outputs = model.generate(input_ids=input_ids, max_length=50, temperature=0.7, repetition_penalty=1.2)  # generate sequences
             print('Generated: {}'.format(tokenizer.decode(outputs[0], skip_special_tokens=True)))
 
             tokenizer = AutoTokenizer.from_pretrained('gpt2')   # Initialize tokenizer
-            model = TFAutoModelWithLMHead.from_pretrained('gpt2')    # Download model and configuration from S3 and cache.
+            model = TFAutoModelWithLMHead.from_pretrained('gpt2')    # Download model and configuration from huggingface.co and cache.
             input_context = 'My cute dog'
             bad_words_ids = [tokenizer.encode(bad_word, add_prefix_space=True) for bad_word in ['idiot', 'stupid', 'shut up']]
             input_ids = tokenizer.encode(input_context, return_tensors='tf')  # encode input context

src/transformers/modelcard.py

@@ -87,10 +87,9 @@ class ModelCard:
         Parameters:
             pretrained_model_name_or_path: either:
 
-                - a string with the `shortcut name` of a pre-trained model card to load from cache or download, e.g.:
+                - a string, the `model id` of a pretrained model card hosted inside a model repo on huggingface.co.
-                  ``bert-base-uncased``.
+                  Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under a
-                - a string with the `identifier name` of a pre-trained model card that was user-uploaded to our S3,
+                  user or organization name, like ``dbmdz/bert-base-german-cased``.
-                  e.g.: ``dbmdz/bert-base-german-cased``.
                 - a path to a `directory` containing a model card file saved using the
                   :func:`~transformers.ModelCard.save_pretrained` method, e.g.: ``./my_model_directory/``.
                 - a path or url to a saved model card JSON `file`, e.g.: ``./my_model_directory/modelcard.json``.
@@ -124,7 +123,7 @@ class ModelCard:
 
         Examples::
 
-            modelcard = ModelCard.from_pretrained('bert-base-uncased')    # Download model card from S3 and cache.
+            modelcard = ModelCard.from_pretrained('bert-base-uncased')    # Download model card from huggingface.co and cache.
             modelcard = ModelCard.from_pretrained('./test/saved_model/')  # E.g. model card was saved using `save_pretrained('./test/saved_model/')`
             modelcard = ModelCard.from_pretrained('./test/saved_model/modelcard.json')
             modelcard = ModelCard.from_pretrained('bert-base-uncased', output_attentions=True, foo=False)

src/transformers/modeling_tf_utils.py

@@ -544,10 +544,9 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin):
             pretrained_model_name_or_path (:obj:`str`, `optional`):
                 Can be either:
 
-                    - A string with the `shortcut name` of a pretrained model to load from cache or download, e.g.,
+                    - A string, the `model id` of a pretrained model hosted inside a model repo on huggingface.co.
-                      ``bert-base-uncased``.
+                      Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under
-                    - A string with the `identifier name` of a pretrained model that was user-uploaded to our S3, e.g.,
+                      a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing model weights saved using
                       :func:`~transformersTF.PreTrainedModel.save_pretrained`, e.g., ``./my_model_directory/``.
                     - A path or url to a `PyTorch state_dict save file` (e.g, ``./pt_model/pytorch_model.bin``). In
@@ -568,8 +567,8 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin):
                 Configuration for the model to use instead of an automatically loaded configuation. Configuration can
                 be automatically loaded when:
 
-                    - The model is a model provided by the library (loaded with the `shortcut name` string of a
+                    - The model is a model provided by the library (loaded with the `model id` string of a pretrained
-                      pretrained model).
+                      model).
                     - The model was saved using :func:`~transformers.TFPreTrainedModel.save_pretrained` and is reloaded
                       by supplying the save directory.
                     - The model is loaded by supplying a local directory as ``pretrained_model_name_or_path`` and a
@@ -618,7 +617,7 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin):
         Examples::
 
             >>> from transformers import BertConfig, TFBertModel
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFBertModel.from_pretrained('bert-base-uncased')
             >>> # Model was saved using `save_pretrained('./test/saved_model/')` (for example purposes, not runnable).
             >>> model = TFBertModel.from_pretrained('./test/saved_model/')

src/transformers/modeling_utils.py

@@ -758,10 +758,9 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin):
             pretrained_model_name_or_path (:obj:`str`, `optional`):
                 Can be either:
 
-                    - A string with the `shortcut name` of a pretrained model to load from cache or download, e.g.,
+                    - A string, the `model id` of a pretrained model hosted inside a model repo on huggingface.co.
-                      ``bert-base-uncased``.
+                      Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under
-                    - A string with the `identifier name` of a pretrained model that was user-uploaded to our S3, e.g.,
+                      a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing model weights saved using
                       :func:`~transformers.PreTrainedModel.save_pretrained`, e.g., ``./my_model_directory/``.
                     - A path or url to a `tensorflow index checkpoint file` (e.g, ``./tf_model/model.ckpt.index``). In
@@ -781,8 +780,8 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin):
                 Configuration for the model to use instead of an automatically loaded configuation. Configuration can
                 be automatically loaded when:
 
-                    - The model is a model provided by the library (loaded with the `shortcut name` string of a
+                    - The model is a model provided by the library (loaded with the `model id` string of a pretrained
-                      pretrained model).
+                      model).
                     - The model was saved using :func:`~transformers.PreTrainedModel.save_pretrained` and is reloaded
                       by supplying the save directory.
                     - The model is loaded by supplying a local directory as ``pretrained_model_name_or_path`` and a
@@ -838,7 +837,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin):
         Examples::
 
             >>> from transformers import BertConfig, BertModel
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = BertModel.from_pretrained('bert-base-uncased')
             >>> # Model was saved using `save_pretrained('./test/saved_model/')` (for example purposes, not runnable).
             >>> model = BertModel.from_pretrained('./test/saved_model/')

src/transformers/models/auto/configuration_auto.py

@@ -277,10 +277,9 @@ class AutoConfig:
             pretrained_model_name_or_path (:obj:`str`):
                 Can be either:
 
-                    - A string with the `shortcut name` of a pretrained model configuration to load from cache or
+                    - A string, the `model id` of a pretrained model configuration hosted inside a model repo on
-                      download, e.g., ``bert-base-uncased``.
+                      huggingface.co. Valid model ids can be located at the root-level, like ``bert-base-uncased``, or
-                    - A string with the `identifier name` of a pretrained model configuration that was user-uploaded to
+                      namespaced under a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      our S3, e.g., ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing a configuration file saved using the
                       :meth:`~transformers.PretrainedConfig.save_pretrained` method, or the
                       :meth:`~transformers.PreTrainedModel.save_pretrained` method, e.g., ``./my_model_directory/``.
@@ -317,10 +316,10 @@ class AutoConfig:
 
             >>> from transformers import AutoConfig
 
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
 
-            >>> # Download configuration from S3 (user-uploaded) and cache.
+            >>> # Download configuration from huggingface.co (user-uploaded) and cache.
             >>> config = AutoConfig.from_pretrained('dbmdz/bert-base-german-cased')
 
             >>> # If configuration file is in a directory (e.g., was saved using `save_pretrained('./test/saved_model/')`).

src/transformers/models/auto/modeling_auto.py

@@ -505,10 +505,9 @@ AUTO_MODEL_PRETRAINED_DOCSTRING = r"""
             pretrained_model_name_or_path:
                 Can be either:
 
-                    - A string with the `shortcut name` of a pretrained model to load from cache or download, e.g.,
+                    - A string, the `model id` of a pretrained model hosted inside a model repo on huggingface.co.
-                      ``bert-base-uncased``.
+                      Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under
-                    - A string with the `identifier name` of a pretrained model that was user-uploaded to our S3, e.g.,
+                      a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing model weights saved using
                       :func:`~transformers.PreTrainedModel.save_pretrained`, e.g., ``./my_model_directory/``.
                     - A path or url to a `tensorflow index checkpoint file` (e.g, ``./tf_model/model.ckpt.index``). In
@@ -521,8 +520,8 @@ AUTO_MODEL_PRETRAINED_DOCSTRING = r"""
                 Configuration for the model to use instead of an automatically loaded configuration. Configuration can
                 be automatically loaded when:
 
-                    - The model is a model provided by the library (loaded with the `shortcut name` string of a
+                    - The model is a model provided by the library (loaded with the `model id` string of a pretrained
-                      pretrained model).
+                      model).
                     - The model was saved using :meth:`~transformers.PreTrainedModel.save_pretrained` and is reloaded
                       by supplying the save directory.
                     - The model is loaded by supplying a local directory as ``pretrained_model_name_or_path`` and a
@@ -608,7 +607,7 @@ class AutoModel:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModel
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = AutoModel.from_config(config)
         """
@@ -634,7 +633,7 @@ class AutoModel:
 
             >>> from transformers import AutoConfig, AutoModel
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModel.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -702,7 +701,7 @@ class AutoModelForPreTraining:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelForPreTraining
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = AutoModelForPreTraining.from_config(config)
         """
@@ -728,7 +727,7 @@ class AutoModelForPreTraining:
 
             >>> from transformers import AutoConfig, AutoModelForPreTraining
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelForPreTraining.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -802,7 +801,7 @@ class AutoModelWithLMHead:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelWithLMHead
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = AutoModelWithLMHead.from_config(config)
         """
@@ -834,7 +833,7 @@ class AutoModelWithLMHead:
 
             >>> from transformers import AutoConfig, AutoModelWithLMHead
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelWithLMHead.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -908,7 +907,7 @@ class AutoModelForCausalLM:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelForCausalLM
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('gpt2')
             >>> model = AutoModelForCausalLM.from_config(config)
         """
@@ -934,7 +933,7 @@ class AutoModelForCausalLM:
 
             >>> from transformers import AutoConfig, AutoModelForCausalLM
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelForCausalLM.from_pretrained('gpt2')
 
             >>> # Update configuration during loading
@@ -1002,7 +1001,7 @@ class AutoModelForMaskedLM:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelForMaskedLM
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = AutoModelForMaskedLM.from_config(config)
         """
@@ -1028,7 +1027,7 @@ class AutoModelForMaskedLM:
 
             >>> from transformers import AutoConfig, AutoModelForMaskedLM
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelForMaskedLM.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -1096,7 +1095,7 @@ class AutoModelForSeq2SeqLM:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelForSeq2SeqLM
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('t5')
             >>> model = AutoModelForSeq2SeqLM.from_config(config)
         """
@@ -1124,7 +1123,7 @@ class AutoModelForSeq2SeqLM:
 
             >>> from transformers import AutoConfig, AutoModelForSeq2SeqLM
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelForSeq2SeqLM.from_pretrained('t5-base')
 
             >>> # Update configuration during loading
@@ -1194,7 +1193,7 @@ class AutoModelForSequenceClassification:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelForSequenceClassification
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = AutoModelForSequenceClassification.from_config(config)
         """
@@ -1222,7 +1221,7 @@ class AutoModelForSequenceClassification:
 
             >>> from transformers import AutoConfig, AutoModelForSequenceClassification
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelForSequenceClassification.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -1291,7 +1290,7 @@ class AutoModelForQuestionAnswering:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelForQuestionAnswering
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = AutoModelForQuestionAnswering.from_config(config)
         """
@@ -1320,7 +1319,7 @@ class AutoModelForQuestionAnswering:
 
             >>> from transformers import AutoConfig, AutoModelForQuestionAnswering
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelForQuestionAnswering.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -1390,7 +1389,7 @@ class AutoModelForTokenClassification:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelForTokenClassification
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = AutoModelForTokenClassification.from_config(config)
         """
@@ -1419,7 +1418,7 @@ class AutoModelForTokenClassification:
 
             >>> from transformers import AutoConfig, AutoModelForTokenClassification
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelForTokenClassification.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -1490,7 +1489,7 @@ class AutoModelForMultipleChoice:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelForMultipleChoice
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = AutoModelForMultipleChoice.from_config(config)
         """
@@ -1519,7 +1518,7 @@ class AutoModelForMultipleChoice:
 
             >>> from transformers import AutoConfig, AutoModelForMultipleChoice
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelForMultipleChoice.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -1590,7 +1589,7 @@ class AutoModelForNextSentencePrediction:
         Examples::
 
             >>> from transformers import AutoConfig, AutoModelForNextSentencePrediction
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = AutoModelForNextSentencePrediction.from_config(config)
         """
@@ -1619,7 +1618,7 @@ class AutoModelForNextSentencePrediction:
 
             >>> from transformers import AutoConfig, AutoModelForNextSentencePrediction
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = AutoModelForNextSentencePrediction.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading

src/transformers/models/auto/modeling_flax_auto.py

@@ -75,7 +75,7 @@ class FlaxAutoModel(object):
         Examples::
 
             config = BertConfig.from_pretrained('bert-base-uncased')
-            # Download configuration from S3 and cache.
+            # Download configuration from huggingface.co and cache.
             model = FlaxAutoModel.from_config(config)
             # E.g. model was saved using `save_pretrained('./test/saved_model/')`
         """
@@ -109,10 +109,9 @@ class FlaxAutoModel(object):
         Args:
             pretrained_model_name_or_path: either:
 
-                - a string with the `shortcut name` of a pre-trained model to load from cache or download, e.g.:
+                - a string, the `model id` of a pretrained model hosted inside a model repo on huggingface.co. Valid
-                  ``bert-base-uncased``.
+                  model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under a user or
-                - a string with the `identifier name` of a pre-trained model that was user-uploaded to our S3, e.g.:
+                  organization name, like ``dbmdz/bert-base-german-cased``.
-                  ``dbmdz/bert-base-german-cased``.
                 - a path to a `directory` containing model weights saved using
                   :func:`~transformers.FlaxPreTrainedModel.save_pretrained`, e.g.: ``./my_model_directory/``.
                 - a path or url to a `tensorflow index checkpoint file` (e.g. `./tf_model/model.ckpt.index`). In this
@@ -165,7 +164,7 @@ class FlaxAutoModel(object):
 
         Examples::
 
-            model = FlaxAutoModel.from_pretrained('bert-base-uncased')    # Download model and configuration from S3 and cache.
+            model = FlaxAutoModel.from_pretrained('bert-base-uncased')    # Download model and configuration from huggingface.co and cache.
             model = FlaxAutoModel.from_pretrained('./test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
             assert model.config.output_attention == True
 

src/transformers/models/auto/modeling_tf_auto.py

@@ -403,10 +403,9 @@ TF_AUTO_MODEL_PRETRAINED_DOCSTRING = r"""
             pretrained_model_name_or_path:
                 Can be either:
 
-                    - A string with the `shortcut name` of a pretrained model to load from cache or download, e.g.,
+                    - A string, the `model id` of a pretrained model hosted inside a model repo on huggingface.co.
-                      ``bert-base-uncased``.
+                      Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under
-                    - A string with the `identifier name` of a pretrained model that was user-uploaded to our S3, e.g.,
+                      a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing model weights saved using
                       :func:`~transformers.PreTrainedModel.save_pretrained`, e.g., ``./my_model_directory/``.
                     - A path or url to a `PyTorch state_dict save file` (e.g, ``./pt_model/pytorch_model.bin``). In
@@ -420,8 +419,8 @@ TF_AUTO_MODEL_PRETRAINED_DOCSTRING = r"""
                 Configuration for the model to use instead of an automatically loaded configuration. Configuration can
                 be automatically loaded when:
 
-                    - The model is a model provided by the library (loaded with the `shortcut name` string of a
+                    - The model is a model provided by the library (loaded with the `model id` string of a pretrained
-                      pretrained model).
+                      model).
                     - The model was saved using :meth:`~transformers.PreTrainedModel.save_pretrained` and is reloaded
                       by suppyling the save directory.
                     - The model is loaded by suppyling a local directory as ``pretrained_model_name_or_path`` and a
@@ -507,7 +506,7 @@ class TFAutoModel(object):
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModel
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = TFAutoConfig.from_pretrained('bert-base-uncased')
             >>> model = TFAutoModel.from_config(config)
         """
@@ -533,7 +532,7 @@ class TFAutoModel(object):
 
             >>> from transformers import AutoConfig, AutoModel
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModel.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -601,7 +600,7 @@ class TFAutoModelForPreTraining(object):
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelForPreTraining
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = TFAutoModelForPreTraining.from_config(config)
         """
@@ -627,7 +626,7 @@ class TFAutoModelForPreTraining(object):
 
             >>> from transformers import AutoConfig, TFAutoModelForPreTraining
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelForPreTraining.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -701,7 +700,7 @@ class TFAutoModelWithLMHead(object):
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelWithLMHead
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = TFAutoModelWithLMHead.from_config(config)
         """
@@ -733,7 +732,7 @@ class TFAutoModelWithLMHead(object):
 
             >>> from transformers import AutoConfig, TFAutoModelWithLMHead
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelWithLMHead.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -808,7 +807,7 @@ class TFAutoModelForCausalLM:
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelForCausalLM
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('gpt2')
             >>> model = TFAutoModelForCausalLM.from_config(config)
         """
@@ -834,7 +833,7 @@ class TFAutoModelForCausalLM:
 
             >>> from transformers import AutoConfig, TFAutoModelForCausalLM
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelForCausalLM.from_pretrained('gpt2')
 
             >>> # Update configuration during loading
@@ -902,7 +901,7 @@ class TFAutoModelForMaskedLM:
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelForMaskedLM
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = TFAutoModelForMaskedLM.from_config(config)
         """
@@ -928,7 +927,7 @@ class TFAutoModelForMaskedLM:
 
             >>> from transformers import AutoConfig, TFAutoModelForMaskedLM
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelForMaskedLM.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -996,7 +995,7 @@ class TFAutoModelForSeq2SeqLM:
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelForSeq2SeqLM
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('t5')
             >>> model = TFAutoModelForSeq2SeqLM.from_config(config)
         """
@@ -1024,7 +1023,7 @@ class TFAutoModelForSeq2SeqLM:
 
             >>> from transformers import AutoConfig, TFAutoModelForSeq2SeqLM
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelForSeq2SeqLM.from_pretrained('t5-base')
 
             >>> # Update configuration during loading
@@ -1094,7 +1093,7 @@ class TFAutoModelForSequenceClassification(object):
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelForSequenceClassification
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = TFAutoModelForSequenceClassification.from_config(config)
         """
@@ -1122,7 +1121,7 @@ class TFAutoModelForSequenceClassification(object):
 
             >>> from transformers import AutoConfig, TFAutoModelForSequenceClassification
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelForSequenceClassification.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -1191,7 +1190,7 @@ class TFAutoModelForQuestionAnswering(object):
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelForQuestionAnswering
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = TFAutoModelForQuestionAnswering.from_config(config)
         """
@@ -1219,7 +1218,7 @@ class TFAutoModelForQuestionAnswering(object):
 
             >>> from transformers import AutoConfig, TFAutoModelForQuestionAnswering
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelForQuestionAnswering.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -1288,7 +1287,7 @@ class TFAutoModelForTokenClassification:
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelForTokenClassification
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = TFAutoModelForTokenClassification.from_config(config)
         """
@@ -1316,7 +1315,7 @@ class TFAutoModelForTokenClassification:
 
             >>> from transformers import AutoConfig, TFAutoModelForTokenClassification
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelForTokenClassification.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -1386,7 +1385,7 @@ class TFAutoModelForMultipleChoice:
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelForMultipleChoice
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = TFAutoModelForMultipleChoice.from_config(config)
         """
@@ -1414,7 +1413,7 @@ class TFAutoModelForMultipleChoice:
 
             >>> from transformers import AutoConfig, TFAutoModelForMultipleChoice
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelForMultipleChoice.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading
@@ -1484,7 +1483,7 @@ class TFAutoModelForNextSentencePrediction:
         Examples::
 
             >>> from transformers import AutoConfig, TFAutoModelForNextSentencePrediction
-            >>> # Download configuration from S3 and cache.
+            >>> # Download configuration from huggingface.co and cache.
             >>> config = AutoConfig.from_pretrained('bert-base-uncased')
             >>> model = TFAutoModelForNextSentencePrediction.from_config(config)
         """
@@ -1512,7 +1511,7 @@ class TFAutoModelForNextSentencePrediction:
 
             >>> from transformers import AutoConfig, TFAutoModelForNextSentencePrediction
 
-            >>> # Download model and configuration from S3 and cache.
+            >>> # Download model and configuration from huggingface.co and cache.
             >>> model = TFAutoModelForNextSentencePrediction.from_pretrained('bert-base-uncased')
 
             >>> # Update configuration during loading

src/transformers/models/auto/tokenization_auto.py

@@ -250,10 +250,9 @@ class AutoTokenizer:
             pretrained_model_name_or_path (:obj:`str`):
                 Can be either:
 
-                    - A string with the `shortcut name` of a predefined tokenizer to load from cache or download, e.g.,
+                    - A string, the `model id` of a predefined tokenizer hosted inside a model repo on huggingface.co.
-                      ``bert-base-uncased``.
+                      Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under
-                    - A string with the `identifier name` of a predefined tokenizer that was user-uploaded to our S3,
+                      a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      e.g., ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing vocabulary files required by the tokenizer, for instance saved
                       using the :func:`~transformers.PreTrainedTokenizer.save_pretrained` method, e.g.,
                       ``./my_model_directory/``.
@@ -280,6 +279,9 @@ class AutoTokenizer:
                 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.
+            subfolder (:obj:`str`, `optional`):
+                In case the relevant files are located inside a subfolder of the model repo on huggingface.co (e.g. for
+                facebook/rag-token-base), specify it here.
             use_fast (:obj:`bool`, `optional`, defaults to :obj:`True`):
                 Whether or not to try to load the fast version of the tokenizer.
             kwargs (additional keyword arguments, `optional`):
@@ -291,10 +293,10 @@ class AutoTokenizer:
 
             >>> from transformers import AutoTokenizer
 
-            >>> # Download vocabulary from S3 and cache.
+            >>> # Download vocabulary from huggingface.co and cache.
             >>> tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
 
-            >>> # Download vocabulary from S3 (user-uploaded) and cache.
+            >>> # Download vocabulary from huggingface.co (user-uploaded) and cache.
             >>> tokenizer = AutoTokenizer.from_pretrained('dbmdz/bert-base-german-cased')
 
             >>> # If vocabulary files are in a directory (e.g. tokenizer was saved using `save_pretrained('./test/saved_model/')`)

src/transformers/models/encoder_decoder/modeling_encoder_decoder.py

@@ -214,10 +214,9 @@ class EncoderDecoderModel(PreTrainedModel):
             encoder_pretrained_model_name_or_path (:obj: `str`, `optional`):
                 Information necessary to initiate the encoder. Can be either:
 
-                    - A string with the `shortcut name` of a pretrained model to load from cache or download, e.g.,
+                    - A string, the `model id` of a pretrained model hosted inside a model repo on huggingface.co.
-                      ``bert-base-uncased``.
+                      Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under
-                    - A string with the `identifier name` of a pretrained model that was user-uploaded to our S3, e.g.,
+                      a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing model weights saved using
                       :func:`~transformers.PreTrainedModel.save_pretrained`, e.g., ``./my_model_directory/``.
                     - A path or url to a `tensorflow index checkpoint file` (e.g, ``./tf_model/model.ckpt.index``). In
@@ -228,10 +227,9 @@ class EncoderDecoderModel(PreTrainedModel):
             decoder_pretrained_model_name_or_path (:obj: `str`, `optional`, defaults to `None`):
                 Information necessary to initiate the decoder. Can be either:
 
-                    - A string with the `shortcut name` of a pretrained model to load from cache or download, e.g.,
+                    - A string, the `model id` of a pretrained model hosted inside a model repo on huggingface.co.
-                      ``bert-base-uncased``.
+                      Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under
-                    - A string with the `identifier name` of a pretrained model that was user-uploaded to our S3, e.g.,
+                      a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing model weights saved using
                       :func:`~transformers.PreTrainedModel.save_pretrained`, e.g., ``./my_model_directory/``.
                     - A path or url to a `tensorflow index checkpoint file` (e.g, ``./tf_model/model.ckpt.index``). In

src/transformers/models/lxmert/tokenization_lxmert.py

@@ -24,7 +24,7 @@ VOCAB_FILES_NAMES = {"vocab_file": "vocab.txt"}
 
 ####################################################
 # Mapping from the keyword arguments names of Tokenizer `__init__`
-# to pretrained vocabulary URL for all the model shortcut names.
+# to pretrained vocabulary URL for all the model ids.
 ####################################################
 PRETRAINED_VOCAB_FILES_MAP = {
     "vocab_file": {
@@ -33,13 +33,13 @@ PRETRAINED_VOCAB_FILES_MAP = {
 }
 
 ####################################################
-# Mapping from model shortcut names to max length of inputs
+# Mapping from model ids to max length of inputs
 ####################################################
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     "unc-nlp/lxmert-base-uncased": 512,
 }
 ####################################################
-# Mapping from model shortcut names to a dictionary of additional
+# Mapping from model ids to a dictionary of additional
 # keyword arguments for Tokenizer `__init__`.
 # To be used for checkpoint specific configurations.
 ####################################################

src/transformers/models/lxmert/tokenization_lxmert_fast.py

@@ -25,7 +25,7 @@ VOCAB_FILES_NAMES = {"vocab_file": "vocab.txt", "tokenizer_file": "tokenizer.jso
 
 ####################################################
 # Mapping from the keyword arguments names of Tokenizer `__init__`
-# to pretrained vocabulary URL for all the model shortcut names.
+# to pretrained vocabulary URL for all the model ids.
 ####################################################
 PRETRAINED_VOCAB_FILES_MAP = {
     "vocab_file": {
@@ -37,13 +37,13 @@ PRETRAINED_VOCAB_FILES_MAP = {
 }
 
 ####################################################
-# Mapping from model shortcut names to max length of inputs
+# Mapping from model ids to max length of inputs
 ####################################################
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     "unc-nlp/lxmert-base-uncased": 512,
 }
 ####################################################
-# Mapping from model shortcut names to a dictionary of additional
+# Mapping from model ids to a dictionary of additional
 # keyword arguments for Tokenizer `__init__`.
 # To be used for checkpoint specific configurations.
 ####################################################

src/transformers/models/rag/modeling_rag.py

@@ -238,10 +238,9 @@ class RagPreTrainedModel(PreTrainedModel):
             question_encoder_pretrained_model_name_or_path (:obj: `str`, `optional`, defaults to `None`):
                 Information necessary to initiate the question encoder. Can be either:
 
-                    - A string with the `shortcut name` of a pretrained model to load from cache or download, e.g.,
+                    - A string, the `model id` of a pretrained model hosted inside a model repo on huggingface.co.
-                      ``bert-base-uncased``.
+                      Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under
-                    - A string with the `identifier name` of a pretrained model that was user-uploaded to our S3, e.g.,
+                      a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing model weights saved using
                       :func:`~transformers.PreTrainedModel.save_pretrained`, e.g., ``./my_model_directory/``.
                     - A path or url to a `tensorflow index checkpoint file` (e.g, ``./tf_model/model.ckpt.index``). In
@@ -252,10 +251,9 @@ class RagPreTrainedModel(PreTrainedModel):
             generator_pretrained_model_name_or_path (:obj: `str`, `optional`, defaults to `None`):
                 Information necessary to initiate the generator. Can be either:
 
-                    - A string with the `shortcut name` of a pretrained model to load from cache or download, e.g.,
+                    - A string, the `model id` of a pretrained model hosted inside a model repo on huggingface.co.
-                      ``bert-base-uncased``.
+                      Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under
-                    - A string with the `identifier name` of a pretrained model that was user-uploaded to our S3, e.g.,
+                      a user or organization name, like ``dbmdz/bert-base-german-cased``.
-                      ``dbmdz/bert-base-german-cased``.
                     - A path to a `directory` containing model weights saved using
                       :func:`~transformers.PreTrainedModel.save_pretrained`, e.g., ``./my_model_directory/``.
                     - A path or url to a `tensorflow index checkpoint file` (e.g, ``./tf_model/model.ckpt.index``). In

src/transformers/models/rag/tokenization_rag.py

@@ -49,10 +49,12 @@ class RagTokenizer:
         if config is None:
             config = RagConfig.from_pretrained(pretrained_model_name_or_path)
 
-        question_encoder_path = os.path.join(pretrained_model_name_or_path, "question_encoder_tokenizer")
+        question_encoder = AutoTokenizer.from_pretrained(
-        generator_path = os.path.join(pretrained_model_name_or_path, "generator_tokenizer")
+            pretrained_model_name_or_path, config=config.question_encoder, subfolder="question_encoder_tokenizer"
-        question_encoder = AutoTokenizer.from_pretrained(question_encoder_path, config=config.question_encoder)
+        )
-        generator = AutoTokenizer.from_pretrained(generator_path, config=config.generator)
+        generator = AutoTokenizer.from_pretrained(
+            pretrained_model_name_or_path, config=config.generator, subfolder="generator_tokenizer"
+        )
         return cls(question_encoder=question_encoder, generator=generator)
 
     def __call__(self, *args, **kwargs):

src/transformers/models/reformer/tokenization_reformer.py

@@ -38,7 +38,7 @@ VOCAB_FILES_NAMES = {"vocab_file": "spiece.model"}
 
 ####################################################
 # Mapping from the keyword arguments names of Tokenizer `__init__`
-# to pretrained vocabulary URL for all the model shortcut names.
+# to pretrained vocabulary URL for all the model ids.
 ####################################################
 PRETRAINED_VOCAB_FILES_MAP = {
     "vocab_file": {
@@ -47,7 +47,7 @@ PRETRAINED_VOCAB_FILES_MAP = {
 }
 
 ####################################################
-# Mapping from model shortcut names to max length of inputs
+# Mapping from model ids to max length of inputs
 ####################################################
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     "google/reformer-crime-and-punishment": 524288,

src/transformers/models/reformer/tokenization_reformer_fast.py

@@ -43,7 +43,7 @@ VOCAB_FILES_NAMES = {"vocab_file": "spiece.model", "tokenizer_file": "tokenizer.
 
 ####################################################
 # Mapping from the keyword arguments names of Tokenizer `__init__`
-# to pretrained vocabulary URL for all the model shortcut names.
+# to pretrained vocabulary URL for all the model ids.
 ####################################################
 PRETRAINED_VOCAB_FILES_MAP = {
     "vocab_file": {
@@ -55,7 +55,7 @@ PRETRAINED_VOCAB_FILES_MAP = {
 }
 
 ####################################################
-# Mapping from model shortcut names to max length of inputs
+# Mapping from model ids to max length of inputs
 ####################################################
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     "google/reformer-crime-and-punishment": 524288,

src/transformers/models/t5/modeling_t5.py

@@ -50,7 +50,7 @@ _CONFIG_FOR_DOC = "T5Config"
 _TOKENIZER_FOR_DOC = "T5Tokenizer"
 
 ####################################################
-# This dict contains shortcut names and associated url
+# This dict contains ids and associated url
 # for the pretrained weights provided with the models
 ####################################################
 T5_PRETRAINED_MODEL_ARCHIVE_LIST = [

src/transformers/models/t5/tokenization_t5.py

@@ -39,7 +39,7 @@ VOCAB_FILES_NAMES = {"vocab_file": "spiece.model"}
 
 ####################################################
 # Mapping from the keyword arguments names of Tokenizer `__init__`
-# to pretrained vocabulary URL for all the model shortcut names.
+# to pretrained vocabulary URL for all the model ids.
 ####################################################
 PRETRAINED_VOCAB_FILES_MAP = {
     "vocab_file": {
@@ -52,7 +52,7 @@ PRETRAINED_VOCAB_FILES_MAP = {
 }
 
 ####################################################
-# Mapping from model shortcut names to max length of inputs
+# Mapping from model ids to max length of inputs
 ####################################################
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     "t5-small": 512,

src/transformers/models/t5/tokenization_t5_fast.py

@@ -42,7 +42,7 @@ VOCAB_FILES_NAMES = {"vocab_file": "spiece.model", "tokenizer_file": "tokenizer.
 
 ####################################################
 # Mapping from the keyword arguments names of Tokenizer `__init__`
-# to pretrained vocabulary URL for all the model shortcut names.
+# to pretrained vocabulary URL for all the model ids.
 ####################################################
 PRETRAINED_VOCAB_FILES_MAP = {
     "vocab_file": {
@@ -62,7 +62,7 @@ PRETRAINED_VOCAB_FILES_MAP = {
 }
 
 ####################################################
-# Mapping from model shortcut names to max length of inputs
+# Mapping from model ids to max length of inputs
 ####################################################
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     "t5-small": 512,

src/transformers/tokenization_utils_base.py

@@ -1615,10 +1615,9 @@ class PreTrainedTokenizerBase(SpecialTokensMixin):
             pretrained_model_name_or_path (:obj:`str`):
                 Can be either:
 
-                - A string with the `shortcut name` of a predefined tokenizer to load from cache or download, e.g.,
+                - A string, the `model id` of a predefined tokenizer hosted inside a model repo on huggingface.co.
-                  ``bert-base-uncased``.
+                  Valid model ids can be located at the root-level, like ``bert-base-uncased``, or namespaced under a
-                - A string with the `identifier name` of a predefined tokenizer that was user-uploaded to our S3, e.g.,
+                  user or organization name, like ``dbmdz/bert-base-german-cased``.
-                  ``dbmdz/bert-base-german-cased``.
                 - A path to a `directory` containing vocabulary files required by the tokenizer, for instance saved
                   using the :meth:`~transformers.tokenization_utils_base.PreTrainedTokenizerBase.save_pretrained`
                   method, e.g., ``./my_model_directory/``.
@@ -1641,6 +1640,9 @@ class PreTrainedTokenizerBase(SpecialTokensMixin):
                 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.
+            subfolder (:obj:`str`, `optional`):
+                In case the relevant files are located inside a subfolder of the model repo on huggingface.co (e.g. for
+                facebook/rag-token-base), specify it here.
             inputs (additional positional arguments, `optional`):
                 Will be passed along to the Tokenizer ``__init__`` method.
             kwargs (additional keyword arguments, `optional`):
@@ -1651,10 +1653,10 @@ class PreTrainedTokenizerBase(SpecialTokensMixin):
         Examples::
 
             # We can't instantiate directly the base class `PreTrainedTokenizerBase` so let's show our examples on a derived class: BertTokenizer
-            # Download vocabulary from S3 and cache.
+            # Download vocabulary from huggingface.co and cache.
             tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
 
-            # Download vocabulary from S3 (user-uploaded) and cache.
+            # Download vocabulary from huggingface.co (user-uploaded) and cache.
             tokenizer = BertTokenizer.from_pretrained('dbmdz/bert-base-german-cased')
 
             # If vocabulary files are in a directory (e.g. tokenizer was saved using `save_pretrained('./test/saved_model/')`)
@@ -1676,6 +1678,7 @@ class PreTrainedTokenizerBase(SpecialTokensMixin):
         proxies = kwargs.pop("proxies", None)
         local_files_only = kwargs.pop("local_files_only", False)
         revision = kwargs.pop("revision", None)
+        subfolder = kwargs.pop("subfolder", None)
 
         s3_models = list(cls.max_model_input_sizes.keys())
         vocab_files = {}
@@ -1722,13 +1725,20 @@ class PreTrainedTokenizerBase(SpecialTokensMixin):
                 # Look for the tokenizer files
                 for file_id, file_name in {**cls.vocab_files_names, **additional_files_names}.items():
                     if os.path.isdir(pretrained_model_name_or_path):
+                        if subfolder is not None:
+                            full_file_name = os.path.join(pretrained_model_name_or_path, subfolder, file_name)
+                        else:
                             full_file_name = os.path.join(pretrained_model_name_or_path, file_name)
                         if not os.path.exists(full_file_name):
                             logger.info("Didn't find file {}. We won't load it.".format(full_file_name))
                             full_file_name = None
                     else:
                         full_file_name = hf_bucket_url(
-                            pretrained_model_name_or_path, filename=file_name, revision=revision, mirror=None
+                            pretrained_model_name_or_path,
+                            filename=file_name,
+                            subfolder=subfolder,
+                            revision=revision,
+                            mirror=None,
                         )
 
                     vocab_files[file_id] = full_file_name

templates/adding_a_new_example_script/{{cookiecutter.directory_name}}/run_{{cookiecutter.example_shortcut}}.py

@@ -75,7 +75,7 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"}
     )
     use_fast_tokenizer: bool = field(
         default=True,
@@ -98,7 +98,7 @@ class ModelArguments:
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
     cache_dir: Optional[str] = field(
-        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"}
     )
     use_fast_tokenizer: bool = field(
         default=True,