Update all references to canonical models (#29001)

* Script & Manual edition

* Update

examples/flax/image-captioning/README.md

@@ -34,7 +34,7 @@ Next, we create a [FlaxVisionEncoderDecoderModel](https://huggingface.co/docs/tr
 python3 create_model_from_encoder_decoder_models.py \
     --output_dir model \
     --encoder_model_name_or_path google/vit-base-patch16-224-in21k \
-    --decoder_model_name_or_path gpt2
+    --decoder_model_name_or_path openai-community/gpt2
 ```
 
 ### Train the model

examples/flax/language-modeling/README.md

@@ -28,7 +28,7 @@ way which enables simple and efficient model parallelism.
 In the following, we demonstrate how to train a bi-directional transformer model 
 using masked language modeling objective as introduced in [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805).
 More specifically, we demonstrate how JAX/Flax can be leveraged 
-to pre-train [**`roberta-base`**](https://huggingface.co/roberta-base)
+to pre-train [**`FacebookAI/roberta-base`**](https://huggingface.co/FacebookAI/roberta-base)
 in Norwegian on a single TPUv3-8 pod.
 
 The example script uses the 🤗 Datasets library. You can easily customize them to your needs if you need extra processing on your datasets.
@@ -76,13 +76,13 @@ tokenizer.save("./norwegian-roberta-base/tokenizer.json")
 ### Create configuration
 
 Next, we create the model's configuration file. This is as simple 
-as loading and storing [`**roberta-base**`](https://huggingface.co/roberta-base)
+as loading and storing [`**FacebookAI/roberta-base**`](https://huggingface.co/FacebookAI/roberta-base)
 in the local model folder:
 
 ```python
 from transformers import RobertaConfig
 
-config = RobertaConfig.from_pretrained("roberta-base", vocab_size=50265)
+config = RobertaConfig.from_pretrained("FacebookAI/roberta-base", vocab_size=50265)
 config.save_pretrained("./norwegian-roberta-base")
 ```
 
@@ -129,8 +129,8 @@ look at [this](https://colab.research.google.com/github/huggingface/notebooks/bl
 
 In the following, we demonstrate how to train an auto-regressive causal transformer model 
 in JAX/Flax.
-More specifically, we pretrain a randomly initialized [**`gpt2`**](https://huggingface.co/gpt2) model in Norwegian on a single TPUv3-8.
-to pre-train 124M [**`gpt2`**](https://huggingface.co/gpt2)
+More specifically, we pretrain a randomly initialized [**`openai-community/gpt2`**](https://huggingface.co/openai-community/gpt2) model in Norwegian on a single TPUv3-8.
+to pre-train 124M [**`openai-community/gpt2`**](https://huggingface.co/openai-community/gpt2)
 in Norwegian on a single TPUv3-8 pod.
 
 The example script uses the 🤗 Datasets library. You can easily customize them to your needs if you need extra processing on your datasets.
@@ -179,13 +179,13 @@ tokenizer.save("./norwegian-gpt2/tokenizer.json")
 ### Create configuration
 
 Next, we create the model's configuration file. This is as simple 
-as loading and storing [`**gpt2**`](https://huggingface.co/gpt2)
+as loading and storing [`**openai-community/gpt2**`](https://huggingface.co/openai-community/gpt2)
 in the local model folder:
 
 ```python
 from transformers import GPT2Config
 
-config = GPT2Config.from_pretrained("gpt2", resid_pdrop=0.0, embd_pdrop=0.0, attn_pdrop=0.0, vocab_size=50257)
+config = GPT2Config.from_pretrained("openai-community/gpt2", resid_pdrop=0.0, embd_pdrop=0.0, attn_pdrop=0.0, vocab_size=50257)
 config.save_pretrained("./norwegian-gpt2")
 ```
 
@@ -199,7 +199,7 @@ Finally, we can run the example script to pretrain the model:
 ```bash
 python run_clm_flax.py \
     --output_dir="./norwegian-gpt2" \
-    --model_type="gpt2" \
+    --model_type="openai-community/gpt2" \
     --config_name="./norwegian-gpt2" \
     --tokenizer_name="./norwegian-gpt2" \
     --dataset_name="oscar" \

examples/flax/question-answering/README.md

@@ -29,7 +29,7 @@ The following example fine-tunes BERT on SQuAD:
 
 ```bash
 python run_qa.py \
-  --model_name_or_path bert-base-uncased \
+  --model_name_or_path google-bert/bert-base-uncased \
   --dataset_name squad \
   --do_train   \
   --do_eval   \
@@ -67,7 +67,7 @@ Here is an example training on 4 TITAN RTX GPUs and Bert Whole Word Masking unca
 ```bash
 export CUDA_VISIBLE_DEVICES=0,1,2,3
 python run_qa.py   \
---model_name_or_path bert-large-uncased-whole-word-masking   \
+--model_name_or_path google-bert/bert-large-uncased-whole-word-masking   \
 --dataset_name squad   \
 --do_train   \
 --do_eval   \

examples/flax/test_flax_examples.py

@@ -78,7 +78,7 @@ class ExamplesTests(TestCasePlus):
         tmp_dir = self.get_auto_remove_tmp_dir()
         testargs = f"""
             run_glue.py
-            --model_name_or_path distilbert-base-uncased
+            --model_name_or_path distilbert/distilbert-base-uncased
             --output_dir {tmp_dir}
             --train_file ./tests/fixtures/tests_samples/MRPC/train.csv
             --validation_file ./tests/fixtures/tests_samples/MRPC/dev.csv
@@ -101,7 +101,7 @@ class ExamplesTests(TestCasePlus):
         tmp_dir = self.get_auto_remove_tmp_dir()
         testargs = f"""
             run_clm_flax.py
-            --model_name_or_path distilgpt2
+            --model_name_or_path distilbert/distilgpt2
             --train_file ./tests/fixtures/sample_text.txt
             --validation_file ./tests/fixtures/sample_text.txt
             --do_train
@@ -125,7 +125,7 @@ class ExamplesTests(TestCasePlus):
         tmp_dir = self.get_auto_remove_tmp_dir()
         testargs = f"""
             run_summarization.py
-            --model_name_or_path t5-small
+            --model_name_or_path google-t5/t5-small
             --train_file tests/fixtures/tests_samples/xsum/sample.json
             --validation_file tests/fixtures/tests_samples/xsum/sample.json
             --test_file tests/fixtures/tests_samples/xsum/sample.json
@@ -155,7 +155,7 @@ class ExamplesTests(TestCasePlus):
         tmp_dir = self.get_auto_remove_tmp_dir()
         testargs = f"""
             run_mlm.py
-            --model_name_or_path distilroberta-base
+            --model_name_or_path distilbert/distilroberta-base
             --train_file ./tests/fixtures/sample_text.txt
             --validation_file ./tests/fixtures/sample_text.txt
             --output_dir {tmp_dir}
@@ -179,7 +179,7 @@ class ExamplesTests(TestCasePlus):
         tmp_dir = self.get_auto_remove_tmp_dir()
         testargs = f"""
             run_t5_mlm_flax.py
-            --model_name_or_path t5-small
+            --model_name_or_path google-t5/t5-small
             --train_file ./tests/fixtures/sample_text.txt
             --validation_file ./tests/fixtures/sample_text.txt
             --do_train
@@ -206,7 +206,7 @@ class ExamplesTests(TestCasePlus):
         tmp_dir = self.get_auto_remove_tmp_dir()
         testargs = f"""
             run_flax_ner.py
-            --model_name_or_path bert-base-uncased
+            --model_name_or_path google-bert/bert-base-uncased
             --train_file tests/fixtures/tests_samples/conll/sample.json
             --validation_file tests/fixtures/tests_samples/conll/sample.json
             --output_dir {tmp_dir}
@@ -233,7 +233,7 @@ class ExamplesTests(TestCasePlus):
         tmp_dir = self.get_auto_remove_tmp_dir()
         testargs = f"""
             run_qa.py
-            --model_name_or_path bert-base-uncased
+            --model_name_or_path google-bert/bert-base-uncased
             --version_2_with_negative
             --train_file tests/fixtures/tests_samples/SQUAD/sample.json
             --validation_file tests/fixtures/tests_samples/SQUAD/sample.json

examples/flax/text-classification/README.md

@@ -31,7 +31,7 @@ GLUE is made up of a total of 9 different tasks. Here is how to run the script o
 export TASK_NAME=mrpc
 
 python run_flax_glue.py \
-  --model_name_or_path bert-base-cased \
+  --model_name_or_path google-bert/bert-base-cased \
   --task_name ${TASK_NAME} \
   --max_seq_length 128 \
   --learning_rate 2e-5 \

examples/flax/token-classification/README.md

@@ -25,7 +25,7 @@ The following example fine-tunes BERT on CoNLL-2003:
 
 ```bash
 python run_flax_ner.py \
-  --model_name_or_path bert-base-cased \
+  --model_name_or_path google-bert/bert-base-cased \
   --dataset_name conll2003 \
   --max_seq_length 128 \
   --learning_rate 2e-5 \

examples/legacy/benchmarking/README.md

@@ -22,5 +22,5 @@ If you would like to list benchmark results on your favorite models of the [mode
 
 | Benchmark description | Results | Environment info |      Author      |
 |:----------|:-------------|:-------------|------:|
-| PyTorch Benchmark on inference for `bert-base-cased` |[memory](https://github.com/patrickvonplaten/files_to_link_to/blob/master/bert_benchmark/inference_memory.csv) | [env](https://github.com/patrickvonplaten/files_to_link_to/blob/master/bert_benchmark/env.csv) | [Partick von Platen](https://github.com/patrickvonplaten) | 
-| PyTorch Benchmark on inference for `bert-base-cased` |[time](https://github.com/patrickvonplaten/files_to_link_to/blob/master/bert_benchmark/inference_time.csv) | [env](https://github.com/patrickvonplaten/files_to_link_to/blob/master/bert_benchmark/env.csv) | [Partick von Platen](https://github.com/patrickvonplaten) | 
+| PyTorch Benchmark on inference for `google-bert/bert-base-cased` |[memory](https://github.com/patrickvonplaten/files_to_link_to/blob/master/bert_benchmark/inference_memory.csv) | [env](https://github.com/patrickvonplaten/files_to_link_to/blob/master/bert_benchmark/env.csv) | [Partick von Platen](https://github.com/patrickvonplaten) | 
+| PyTorch Benchmark on inference for `google-bert/bert-base-cased` |[time](https://github.com/patrickvonplaten/files_to_link_to/blob/master/bert_benchmark/inference_time.csv) | [env](https://github.com/patrickvonplaten/files_to_link_to/blob/master/bert_benchmark/env.csv) | [Partick von Platen](https://github.com/patrickvonplaten) | 

examples/legacy/question-answering/README.md

 #### Fine-tuning BERT on SQuAD1.0 with relative position embeddings
 
 The following examples show how to fine-tune BERT models with different relative position embeddings. The BERT model 
-`bert-base-uncased` was pretrained with default absolute position embeddings. We provide the following pretrained 
+`google-bert/bert-base-uncased` was pretrained with default absolute position embeddings. We provide the following pretrained 
 models which were pre-trained on the same training data (BooksCorpus and English Wikipedia) as in the BERT model 
 training, but with different relative position embeddings. 
 
@@ -10,7 +10,7 @@ Shaw et al., [Self-Attention with Relative Position Representations](https://arx
 * `zhiheng-huang/bert-base-uncased-embedding-relative-key-query`, trained from scratch with relative embedding method 4 
 in Huang et al. [Improve Transformer Models with Better Relative Position Embeddings](https://arxiv.org/abs/2009.13658)
 * `zhiheng-huang/bert-large-uncased-whole-word-masking-embedding-relative-key-query`, fine-tuned from model 
-`bert-large-uncased-whole-word-masking` with 3 additional epochs with relative embedding method 4 in Huang et al. 
+`google-bert/bert-large-uncased-whole-word-masking` with 3 additional epochs with relative embedding method 4 in Huang et al. 
 [Improve Transformer Models with Better Relative Position Embeddings](https://arxiv.org/abs/2009.13658)
 
 
@@ -61,7 +61,7 @@ torchrun --nproc_per_node=8 ./examples/question-answering/run_squad.py \
     --gradient_accumulation_steps 3
 ```
 Training with the above command leads to the f1 score of 93.52, which is slightly better than the f1 score of 93.15 for 
-`bert-large-uncased-whole-word-masking`.
+`google-bert/bert-large-uncased-whole-word-masking`.
 
 #### Distributed training
 
@@ -69,7 +69,7 @@ Here is an example using distributed training on 8 V100 GPUs and Bert Whole Word
 
 ```bash
 torchrun --nproc_per_node=8 ./examples/question-answering/run_squad.py \
-    --model_name_or_path bert-large-uncased-whole-word-masking \
+    --model_name_or_path google-bert/bert-large-uncased-whole-word-masking \
     --dataset_name squad \
     --do_train \
     --do_eval \
@@ -90,7 +90,7 @@ exact_match = 86.91
 ```
 
 This fine-tuned model is available as a checkpoint under the reference
-[`bert-large-uncased-whole-word-masking-finetuned-squad`](https://huggingface.co/bert-large-uncased-whole-word-masking-finetuned-squad).
+[`google-bert/bert-large-uncased-whole-word-masking-finetuned-squad`](https://huggingface.co/google-bert/bert-large-uncased-whole-word-masking-finetuned-squad).
 
 ## Results
 

examples/legacy/run_camembert.py

@@ -39,8 +39,8 @@ def fill_mask(masked_input, model, tokenizer, topk=5):
     return topk_filled_outputs
 
 
-tokenizer = CamembertTokenizer.from_pretrained("camembert-base")
-model = CamembertForMaskedLM.from_pretrained("camembert-base")
+tokenizer = CamembertTokenizer.from_pretrained("almanach/camembert-base")
+model = CamembertForMaskedLM.from_pretrained("almanach/camembert-base")
 model.eval()
 
 masked_input = "Le camembert est <mask> :)"

examples/legacy/run_openai_gpt.py

@@ -20,7 +20,7 @@
 
     This script with default values fine-tunes and evaluate a pretrained OpenAI GPT on the RocStories dataset:
         python run_openai_gpt.py \
-          --model_name openai-gpt \
+          --model_name openai-community/openai-gpt \
           --do_train \
           --do_eval \
           --train_dataset "$ROC_STORIES_DIR/cloze_test_val__spring2016 - cloze_test_ALL_val.csv" \
@@ -104,7 +104,7 @@ def pre_process_datasets(encoded_datasets, input_len, cap_length, start_token, d
 
 def main():
     parser = argparse.ArgumentParser()
-    parser.add_argument("--model_name", type=str, default="openai-gpt", help="pretrained model name")
+    parser.add_argument("--model_name", type=str, default="openai-community/openai-gpt", help="pretrained model name")
     parser.add_argument("--do_train", action="store_true", help="Whether to run training.")
     parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.")
     parser.add_argument(

examples/legacy/run_transfo_xl.py

@@ -40,7 +40,7 @@ logger = logging.getLogger(__name__)
 
 def main():
     parser = argparse.ArgumentParser(description="PyTorch Transformer Language Model")
-    parser.add_argument("--model_name", type=str, default="transfo-xl-wt103", help="pretrained model name")
+    parser.add_argument("--model_name", type=str, default="transfo-xl/transfo-xl-wt103", help="pretrained model name")
     parser.add_argument(
         "--split", type=str, default="test", choices=["all", "valid", "test"], help="which split to evaluate"
     )

examples/legacy/seq2seq/README.md

@@ -170,7 +170,7 @@ If 'translation' is in your task name, the computed metric will be BLEU. Otherwi
 For t5, you need to specify --task translation_{src}_to_{tgt} as follows:
 ```bash
 export DATA_DIR=wmt_en_ro
-./run_eval.py t5-base \
+./run_eval.py google-t5/t5-base \
     $DATA_DIR/val.source t5_val_generations.txt \
     --reference_path $DATA_DIR/val.target \
     --score_path enro_bleu.json \

examples/legacy/seq2seq/old_test_datasets.py

@@ -28,7 +28,7 @@ from transformers.testing_utils import TestCasePlus, slow
 from utils import FAIRSEQ_AVAILABLE, DistributedSortishSampler, LegacySeq2SeqDataset, Seq2SeqDataset
 
 
-BERT_BASE_CASED = "bert-base-cased"
+BERT_BASE_CASED = "google-bert/bert-base-cased"
 PEGASUS_XSUM = "google/pegasus-xsum"
 ARTICLES = [" Sam ate lunch today.", "Sams lunch ingredients."]
 SUMMARIES = ["A very interesting story about what I ate for lunch.", "Avocado, celery, turkey, coffee"]

examples/legacy/seq2seq/pack_dataset.py

@@ -74,7 +74,7 @@ def pack_data_dir(tok, data_dir: Path, max_tokens, save_path):
 
 def packer_cli():
     parser = argparse.ArgumentParser()
-    parser.add_argument("--tok_name", type=str, help="like facebook/bart-large-cnn,t5-base, etc.")
+    parser.add_argument("--tok_name", type=str, help="like facebook/bart-large-cnn,google-t5/t5-base, etc.")
     parser.add_argument("--max_seq_len", type=int, default=128)
     parser.add_argument("--data_dir", type=str)
     parser.add_argument("--save_path", type=str)

examples/legacy/seq2seq/run_distributed_eval.py

@@ -124,7 +124,7 @@ def run_generate():
     parser.add_argument(
         "--model_name",
         type=str,
-        help="like facebook/bart-large-cnn,t5-base, etc.",
+        help="like facebook/bart-large-cnn,google-t5/t5-base, etc.",
         default="sshleifer/distilbart-xsum-12-3",
     )
     parser.add_argument("--save_dir", type=str, help="where to save", default="tmp_gen")

examples/legacy/seq2seq/run_eval.py

@@ -100,7 +100,7 @@ def run_generate(verbose=True):
     """
 
     parser = argparse.ArgumentParser()
-    parser.add_argument("model_name", type=str, help="like facebook/bart-large-cnn,t5-base, etc.")
+    parser.add_argument("model_name", type=str, help="like facebook/bart-large-cnn,google-t5/t5-base, etc.")
     parser.add_argument("input_path", type=str, help="like cnn_dm/test.source")
     parser.add_argument("save_path", type=str, help="where to save summaries")
     parser.add_argument("--reference_path", type=str, required=False, help="like cnn_dm/test.target")

examples/legacy/token-classification/README.md

@@ -34,7 +34,7 @@ Let's define some variables that we need for further pre-processing steps and tr
 
 ```bash
 export MAX_LENGTH=128
-export BERT_MODEL=bert-base-multilingual-cased
+export BERT_MODEL=google-bert/bert-base-multilingual-cased
 ```
 
 Run the pre-processing script on training, dev and test datasets:
@@ -92,7 +92,7 @@ Instead of passing all parameters via commandline arguments, the `run_ner.py` sc
 {
     "data_dir": ".",
     "labels": "./labels.txt",
-    "model_name_or_path": "bert-base-multilingual-cased",
+    "model_name_or_path": "google-bert/bert-base-multilingual-cased",
     "output_dir": "germeval-model",
     "max_seq_length": 128,
     "num_train_epochs": 3,
@@ -222,7 +222,7 @@ Let's define some variables that we need for further pre-processing steps:
 
 ```bash
 export MAX_LENGTH=128
-export BERT_MODEL=bert-large-cased
+export BERT_MODEL=google-bert/bert-large-cased
 ```
 
 Here we use the English BERT large model for fine-tuning.
@@ -250,7 +250,7 @@ This configuration file looks like:
 {
     "data_dir": "./data_wnut_17",
     "labels": "./data_wnut_17/labels.txt",
-    "model_name_or_path": "bert-large-cased",
+    "model_name_or_path": "google-bert/bert-large-cased",
     "output_dir": "wnut-17-model-1",
     "max_seq_length": 128,
     "num_train_epochs": 3,

examples/legacy/token-classification/utils_ner.py

@@ -113,7 +113,7 @@ class TokenClassificationTask:
             for word, label in zip(example.words, example.labels):
                 word_tokens = tokenizer.tokenize(word)
 
-                # bert-base-multilingual-cased sometimes output "nothing ([]) when calling tokenize with just a space.
+                # google-bert/bert-base-multilingual-cased sometimes output "nothing ([]) when calling tokenize with just a space.
                 if len(word_tokens) > 0:
                     tokens.extend(word_tokens)
                     # Use the real label id for the first token of the word, and padding ids for the remaining tokens

examples/pytorch/README.md

@@ -109,7 +109,7 @@ classification MNLI task using the `run_glue` script, with 8 GPUs:
 ```bash
 torchrun \
     --nproc_per_node 8 pytorch/text-classification/run_glue.py \
-    --model_name_or_path bert-large-uncased-whole-word-masking \
+    --model_name_or_path google-bert/bert-large-uncased-whole-word-masking \
     --task_name mnli \
     --do_train \
     --do_eval \
@@ -153,7 +153,7 @@ classification MNLI task using the `run_glue` script, with 8 TPUs (from this fol
 ```bash
 python xla_spawn.py --num_cores 8 \
     text-classification/run_glue.py \
-    --model_name_or_path bert-large-uncased-whole-word-masking \
+    --model_name_or_path google-bert/bert-large-uncased-whole-word-masking \
     --task_name mnli \
     --do_train \
     --do_eval \

examples/pytorch/contrastive-image-text/README.md

@@ -64,10 +64,10 @@ from transformers import (
 )
 
 model = VisionTextDualEncoderModel.from_vision_text_pretrained(
-    "openai/clip-vit-base-patch32", "roberta-base"
+    "openai/clip-vit-base-patch32", "FacebookAI/roberta-base"
 )
 
-tokenizer = AutoTokenizer.from_pretrained("roberta-base")
+tokenizer = AutoTokenizer.from_pretrained("FacebookAI/roberta-base")
 image_processor = AutoImageProcessor.from_pretrained("openai/clip-vit-base-patch32")
 processor = VisionTextDualEncoderProcessor(image_processor, tokenizer)