Update legacy Repository usage in various example files (#29085)

* Update legacy Repository usage in `examples/pytorch/text-classification/run_glue_no_trainer.py`

Marked for deprecation here https://huggingface.co/docs/huggingface_hub/guides/upload#legacy-upload-files-with-git-lfs

* Fix import order

* Replace all example usage of deprecated Repository

* Fix remaining repo call and rename args variable

* Revert removing creation of gitignore files and don't change research examples

examples/flax/image-captioning/run_image_captioning_flax.py

@@ -42,7 +42,7 @@ from flax import jax_utils, traverse_util
 from flax.jax_utils import unreplicate
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard, shard_prng_key
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from PIL import Image
 from tqdm import tqdm
 
@@ -455,9 +455,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # Get the datasets: you can either provide your own CSV/JSON training and evaluation files (see below)
     # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
@@ -1061,7 +1060,13 @@ def main():
             model.save_pretrained(os.path.join(training_args.output_dir, ckpt_dir), params=params)
             tokenizer.save_pretrained(os.path.join(training_args.output_dir, ckpt_dir))
             if training_args.push_to_hub:
-                repo.push_to_hub(commit_message=commit_msg, blocking=False)
+                api.upload_folder(
+                    commit_message=commit_msg,
+                    folder_path=training_args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=training_args.hub_token,
+                )
 
     def evaluation_loop(
         rng: jax.random.PRNGKey,

examples/flax/language-modeling/run_bart_dlm_flax.py

@@ -44,7 +44,7 @@ from flax import jax_utils, traverse_util
 from flax.jax_utils import pad_shard_unpad
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from tqdm import tqdm
 
 from transformers import (
@@ -517,9 +517,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
     # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
@@ -949,7 +948,13 @@ def main():
                     model.save_pretrained(training_args.output_dir, params=params)
                     tokenizer.save_pretrained(training_args.output_dir)
                     if training_args.push_to_hub:
-                        repo.push_to_hub(commit_message=f"Saving weights and logs of step {cur_step}", blocking=False)
+                        api.upload_folder(
+                            commit_message=f"Saving weights and logs of step {cur_step}",
+                            folder_path=training_args.output_dir,
+                            repo_id=repo_id,
+                            repo_type="model",
+                            token=training_args.hub_token,
+                        )
 
     # Eval after training
     if training_args.do_eval:

examples/flax/language-modeling/run_clm_flax.py

@@ -44,7 +44,7 @@ from flax import jax_utils, traverse_util
 from flax.jax_utils import pad_shard_unpad, unreplicate
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard, shard_prng_key
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from tqdm import tqdm
 
 import transformers
@@ -403,9 +403,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     #  Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
     # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
@@ -847,8 +846,13 @@ def main():
                     model.save_pretrained(training_args.output_dir, params=params)
                     tokenizer.save_pretrained(training_args.output_dir)
                     if training_args.push_to_hub:
-                        repo.push_to_hub(commit_message=f"Saving weights and logs of step {cur_step}", blocking=False)
-
+                        api.upload_folder(
+                            commit_message=f"Saving weights and logs of step {cur_step}",
+                            folder_path=training_args.output_dir,
+                            repo_id=repo_id,
+                            repo_type="model",
+                            token=training_args.hub_token,
+                        )
     # Eval after training
     if training_args.do_eval:
         eval_metrics = []

examples/flax/language-modeling/run_mlm_flax.py

@@ -45,7 +45,7 @@ from flax import jax_utils, traverse_util
 from flax.jax_utils import pad_shard_unpad
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from tqdm import tqdm
 
 from transformers import (
@@ -441,9 +441,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
     # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
@@ -890,8 +889,13 @@ def main():
                     model.save_pretrained(training_args.output_dir, params=params)
                     tokenizer.save_pretrained(training_args.output_dir)
                     if training_args.push_to_hub:
-                        repo.push_to_hub(commit_message=f"Saving weights and logs of step {cur_step}", blocking=False)
-
+                        api.upload_folder(
+                            commit_message=f"Saving weights and logs of step {cur_step}",
+                            folder_path=training_args.output_dir,
+                            repo_id=repo_id,
+                            repo_type="model",
+                            token=training_args.hub_token,
+                        )
     # Eval after training
     if training_args.do_eval:
         num_eval_samples = len(tokenized_datasets["validation"])

examples/flax/language-modeling/run_t5_mlm_flax.py

@@ -44,7 +44,7 @@ from flax import jax_utils, traverse_util
 from flax.jax_utils import pad_shard_unpad
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from tqdm import tqdm
 
 from transformers import (
@@ -558,9 +558,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
     # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
@@ -977,8 +976,13 @@ def main():
                     model.save_pretrained(training_args.output_dir, params=params)
                     tokenizer.save_pretrained(training_args.output_dir)
                     if training_args.push_to_hub:
-                        repo.push_to_hub(commit_message=f"Saving weights and logs of step {cur_step}", blocking=False)
-
+                        api.upload_folder(
+                            commit_message=f"Saving weights and logs of step {cur_step}",
+                            folder_path=training_args.output_dir,
+                            repo_id=repo_id,
+                            repo_type="model",
+                            token=training_args.hub_token,
+                        )
     # Eval after training
     if training_args.do_eval:
         num_eval_samples = len(tokenized_datasets["validation"])

examples/flax/question-answering/run_qa.py

@@ -42,7 +42,7 @@ from flax import struct, traverse_util
 from flax.jax_utils import pad_shard_unpad, replicate, unreplicate
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from tqdm import tqdm
 from utils_qa import postprocess_qa_predictions
 
@@ -493,9 +493,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # region Load Data
     # Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
@@ -1051,7 +1050,13 @@ def main():
                     model.save_pretrained(training_args.output_dir, params=params)
                     tokenizer.save_pretrained(training_args.output_dir)
                     if training_args.push_to_hub:
-                        repo.push_to_hub(commit_message=f"Saving weights and logs of step {cur_step}", blocking=False)
+                        api.upload_folder(
+                            commit_message=f"Saving weights and logs of step {cur_step}",
+                            folder_path=training_args.output_dir,
+                            repo_id=repo_id,
+                            repo_type="model",
+                            token=training_args.hub_token,
+                        )
         epochs.desc = f"Epoch ... {epoch + 1}/{num_epochs}"
     # endregion
 

examples/flax/speech-recognition/run_flax_speech_recognition_seq2seq.py

@@ -39,7 +39,7 @@ from flax import jax_utils, traverse_util
 from flax.jax_utils import pad_shard_unpad, unreplicate
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard, shard_prng_key
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from tqdm import tqdm
 
@@ -427,8 +427,9 @@ def main():
             )
         else:
             repo_name = training_args.hub_model_id
-        create_repo(repo_name, exist_ok=True, token=training_args.hub_token)
-        repo = Repository(training_args.output_dir, clone_from=repo_name, token=training_args.hub_token)
+        # Create repo and retrieve repo_id
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # 3. Load dataset
     raw_datasets = DatasetDict()
@@ -852,7 +853,13 @@ def main():
             model.save_pretrained(training_args.output_dir, params=params)
             tokenizer.save_pretrained(training_args.output_dir)
             if training_args.push_to_hub:
-                repo.push_to_hub(commit_message=f"Saving weights and logs of epoch {epoch}", blocking=False)
+                api.upload_folder(
+                    commit_message=f"Saving weights and logs of epoch {epoch}",
+                    folder_path=training_args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=training_args.hub_token,
+                )
 
 
 if __name__ == "__main__":

examples/flax/summarization/run_summarization_flax.py

@@ -44,7 +44,7 @@ from flax import jax_utils, traverse_util
 from flax.jax_utils import pad_shard_unpad, unreplicate
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard, shard_prng_key
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from tqdm import tqdm
 
 import transformers
@@ -483,9 +483,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # Get the datasets: you can either provide your own CSV/JSON training and evaluation files (see below)
     # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
@@ -976,7 +975,13 @@ def main():
             model.save_pretrained(training_args.output_dir, params=params)
             tokenizer.save_pretrained(training_args.output_dir)
             if training_args.push_to_hub:
-                repo.push_to_hub(commit_message=f"Saving weights and logs of epoch {epoch}", blocking=False)
+                api.upload_folder(
+                    commit_message=f"Saving weights and logs of epoch {epoch}",
+                    folder_path=training_args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=training_args.hub_token,
+                )
 
     # ======================== Prediction loop ==============================
     if training_args.do_predict:

examples/flax/text-classification/run_flax_glue.py

@@ -37,7 +37,7 @@ from flax import struct, traverse_util
 from flax.jax_utils import pad_shard_unpad, replicate, unreplicate
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from tqdm import tqdm
 
 import transformers
@@ -373,9 +373,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # Get the datasets: you can either provide your own CSV/JSON training and evaluation files (see below)
     # or specify a GLUE benchmark task (the dataset will be downloaded automatically from the datasets Hub).
@@ -677,7 +676,13 @@ def main():
                     model.save_pretrained(training_args.output_dir, params=params)
                     tokenizer.save_pretrained(training_args.output_dir)
                     if training_args.push_to_hub:
-                        repo.push_to_hub(commit_message=f"Saving weights and logs of step {cur_step}", blocking=False)
+                        api.upload_folder(
+                            commit_message=f"Saving weights and logs of epoch {epoch}",
+                            folder_path=training_args.output_dir,
+                            repo_id=repo_id,
+                            repo_type="model",
+                            token=training_args.hub_token,
+                        )
             epochs.desc = f"Epoch ... {epoch + 1}/{num_epochs}"
 
     # save the eval metrics in json

examples/flax/token-classification/run_flax_ner.py

@@ -39,7 +39,7 @@ from flax import struct, traverse_util
 from flax.jax_utils import pad_shard_unpad, replicate, unreplicate
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from tqdm import tqdm
 
 import transformers
@@ -429,9 +429,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
     # or just provide the name of one of the public datasets for token classification task available on the hub at https://huggingface.co/datasets/
@@ -798,7 +797,13 @@ def main():
                     model.save_pretrained(training_args.output_dir, params=params)
                     tokenizer.save_pretrained(training_args.output_dir)
                     if training_args.push_to_hub:
-                        repo.push_to_hub(commit_message=f"Saving weights and logs of step {cur_step}", blocking=False)
+                        api.upload_folder(
+                            commit_message=f"Saving weights and logs of step {cur_step}",
+                            folder_path=training_args.output_dir,
+                            repo_id=repo_id,
+                            repo_type="model",
+                            token=training_args.hub_token,
+                        )
         epochs.desc = f"Epoch ... {epoch + 1}/{num_epochs}"
 
     # Eval after training

examples/flax/vision/run_image_classification.py

@@ -42,7 +42,7 @@ from flax import jax_utils
 from flax.jax_utils import pad_shard_unpad, unreplicate
 from flax.training import train_state
 from flax.training.common_utils import get_metrics, onehot, shard, shard_prng_key
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from tqdm import tqdm
 
 import transformers
@@ -324,9 +324,8 @@ def main():
         if repo_name is None:
             repo_name = Path(training_args.output_dir).absolute().name
         # Create repo and retrieve repo_id
-        repo_id = create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
-        # Clone repo locally
-        repo = Repository(training_args.output_dir, clone_from=repo_id, token=training_args.hub_token)
+        api = HfApi()
+        repo_id = api.create_repo(repo_name, exist_ok=True, token=training_args.hub_token).repo_id
 
     # Initialize datasets and pre-processing transforms
     # We use torchvision here for faster pre-processing
@@ -595,7 +594,13 @@ def main():
             params = jax.device_get(jax.tree_util.tree_map(lambda x: x[0], state.params))
             model.save_pretrained(training_args.output_dir, params=params)
             if training_args.push_to_hub:
-                repo.push_to_hub(commit_message=f"Saving weights and logs of epoch {epoch}", blocking=False)
+                api.upload_folder(
+                    commit_message=f"Saving weights and logs of epoch {epoch}",
+                    folder_path=training_args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=training_args.hub_token,
+                )
 
 
 if __name__ == "__main__":

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -27,7 +27,7 @@ from accelerate import Accelerator
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from torchvision.transforms import (
     CenterCrop,
@@ -264,9 +264,8 @@ def main():
             if repo_name is None:
                 repo_name = Path(args.output_dir).absolute().name
             # Create repo and retrieve repo_id
-            repo_id = create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
-            # Clone repo locally
-            repo = Repository(args.output_dir, clone_from=repo_id, token=args.hub_token)
+            api = HfApi()
+            repo_id = api.create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
 
             with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                 if "step_*" not in gitignore:
@@ -561,10 +560,12 @@ def main():
                         )
                         if accelerator.is_main_process:
                             image_processor.save_pretrained(args.output_dir)
-                            repo.push_to_hub(
-                                commit_message=f"Training in progress {completed_steps} steps",
-                                blocking=False,
-                                auto_lfs_prune=True,
+                            api.upload_folder(
+                                commit_message=f"Training in progress epoch {epoch}",
+                                folder_path=args.output_dir,
+                                repo_id=repo_id,
+                                repo_type="model",
+                                token=args.hub_token,
                             )
 
             if completed_steps >= args.max_train_steps:
@@ -603,8 +604,12 @@ def main():
             )
             if accelerator.is_main_process:
                 image_processor.save_pretrained(args.output_dir)
-                repo.push_to_hub(
-                    commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True
+                api.upload_folder(
+                    commit_message=f"Training in progress epoch {epoch}",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
                 )
 
         if args.checkpointing_steps == "epoch":
@@ -625,8 +630,13 @@ def main():
         if accelerator.is_main_process:
             image_processor.save_pretrained(args.output_dir)
             if args.push_to_hub:
-                repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
-
+                api.upload_folder(
+                    commit_message="End of training",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
+                )
             all_results = {f"eval_{k}": v for k, v in eval_metric.items()}
             with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
                 json.dump(all_results, f)

examples/pytorch/image-pretraining/run_mim_no_trainer.py

@@ -26,7 +26,7 @@ import torch
 from accelerate import Accelerator, DistributedType
 from accelerate.utils import set_seed
 from datasets import load_dataset
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from torchvision.transforms import Compose, Lambda, Normalize, RandomHorizontalFlip, RandomResizedCrop, ToTensor
 from tqdm.auto import tqdm
@@ -437,15 +437,15 @@ def main():
             if repo_name is None:
                 repo_name = Path(args.output_dir).absolute().name
             # Create repo and retrieve repo_id
-            repo_id = create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
-            # Clone repo locally
-            repo = Repository(args.output_dir, clone_from=repo_id, token=args.hub_token)
+            api = HfApi()
+            repo_id = api.create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
 
             with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                 if "step_*" not in gitignore:
                     gitignore.write("step_*\n")
                 if "epoch_*" not in gitignore:
                     gitignore.write("epoch_*\n")
+
         elif args.output_dir is not None:
             os.makedirs(args.output_dir, exist_ok=True)
     accelerator.wait_for_everyone()
@@ -781,8 +781,12 @@ def main():
             )
             if accelerator.is_main_process:
                 image_processor.save_pretrained(args.output_dir)
-                repo.push_to_hub(
-                    commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True
+                api.upload_folder(
+                    commit_message=f"Training in progress epoch {epoch}",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
                 )
 
         if args.checkpointing_steps == "epoch":
@@ -803,7 +807,13 @@ def main():
         if accelerator.is_main_process:
             image_processor.save_pretrained(args.output_dir)
             if args.push_to_hub:
-                repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
+                api.upload_folder(
+                    commit_message="End of training",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
+                )
 
 
 if __name__ == "__main__":

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -37,7 +37,7 @@ from accelerate import Accelerator, DistributedType
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
@@ -304,9 +304,8 @@ def main():
             if repo_name is None:
                 repo_name = Path(args.output_dir).absolute().name
             # Create repo and retrieve repo_id
-            repo_id = create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
-            # Clone repo locally
-            repo = Repository(args.output_dir, clone_from=repo_id, token=args.hub_token)
+            api = HfApi()
+            repo_id = api.create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
 
             with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                 if "step_*" not in gitignore:
@@ -682,8 +681,12 @@ def main():
             )
             if accelerator.is_main_process:
                 tokenizer.save_pretrained(args.output_dir)
-                repo.push_to_hub(
-                    commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True
+                api.upload_folder(
+                    commit_message=f"Training in progress epoch {epoch}",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
                 )
 
         if args.checkpointing_steps == "epoch":
@@ -704,8 +707,13 @@ def main():
         if accelerator.is_main_process:
             tokenizer.save_pretrained(args.output_dir)
             if args.push_to_hub:
-                repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
-
+                api.upload_folder(
+                    commit_message="End of training",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
+                )
             with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
                 json.dump({"perplexity": perplexity}, f)
 

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -37,7 +37,7 @@ from accelerate import Accelerator, DistributedType
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
@@ -311,9 +311,8 @@ def main():
             if repo_name is None:
                 repo_name = Path(args.output_dir).absolute().name
             # Create repo and retrieve repo_id
-            repo_id = create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
-            # Clone repo locally
-            repo = Repository(args.output_dir, clone_from=repo_id, token=args.hub_token)
+            api = HfApi()
+            repo_id = api.create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
 
             with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                 if "step_*" not in gitignore:
@@ -720,8 +719,12 @@ def main():
             )
             if accelerator.is_main_process:
                 tokenizer.save_pretrained(args.output_dir)
-                repo.push_to_hub(
-                    commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True
+                api.upload_folder(
+                    commit_message=f"Training in progress epoch {epoch}",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
                 )
 
         if args.checkpointing_steps == "epoch":
@@ -742,8 +745,13 @@ def main():
         if accelerator.is_main_process:
             tokenizer.save_pretrained(args.output_dir)
             if args.push_to_hub:
-                repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
-
+                api.upload_folder(
+                    commit_message="End of training",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
+                )
             with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
                 json.dump({"perplexity": perplexity}, f)
 

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -36,7 +36,7 @@ from accelerate import Accelerator
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
@@ -328,9 +328,8 @@ def main():
             if repo_name is None:
                 repo_name = Path(args.output_dir).absolute().name
             # Create repo and retrieve repo_id
-            repo_id = create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
-            # Clone repo locally
-            repo = Repository(args.output_dir, clone_from=repo_id, token=args.hub_token)
+            api = HfApi()
+            repo_id = api.create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
 
             with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                 if "step_*" not in gitignore:
@@ -661,8 +660,12 @@ def main():
             )
             if accelerator.is_main_process:
                 tokenizer.save_pretrained(args.output_dir)
-                repo.push_to_hub(
-                    commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True
+                api.upload_folder(
+                    commit_message=f"Training in progress epoch {epoch}",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
                 )
 
         if args.checkpointing_steps == "epoch":
@@ -683,8 +686,13 @@ def main():
         if accelerator.is_main_process:
             tokenizer.save_pretrained(args.output_dir)
             if args.push_to_hub:
-                repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
-
+                api.upload_folder(
+                    commit_message="End of training",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
+                )
             all_results = {f"eval_{k}": v for k, v in eval_metric.items()}
             with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
                 json.dump(all_results, f)

examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py

@@ -34,7 +34,7 @@ from accelerate import Accelerator
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 from utils_qa import postprocess_qa_predictions_with_beam_search
@@ -333,9 +333,8 @@ def main():
             if repo_name is None:
                 repo_name = Path(args.output_dir).absolute().name
             # Create repo and retrieve repo_id
-            repo_id = create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
-            # Clone repo locally
-            repo = Repository(args.output_dir, clone_from=repo_id, token=args.hub_token)
+            api = HfApi()
+            repo_id = api.create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
 
             with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                 if "step_*" not in gitignore:
@@ -873,8 +872,12 @@ def main():
             )
             if accelerator.is_main_process:
                 tokenizer.save_pretrained(args.output_dir)
-                repo.push_to_hub(
-                    commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True
+                api.upload_folder(
+                    commit_message=f"Training in progress epoch {epoch}",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
                 )
 
     # initialize all lists to collect the batches
@@ -1020,7 +1023,13 @@ def main():
         if accelerator.is_main_process:
             tokenizer.save_pretrained(args.output_dir)
             if args.push_to_hub:
-                repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
+                api.upload_folder(
+                    commit_message="End of training",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
+                )
 
             logger.info(json.dumps(eval_metric, indent=4))
             save_prefixed_metrics(eval_metric, args.output_dir)

examples/pytorch/question-answering/run_qa_no_trainer.py

@@ -34,7 +34,7 @@ from accelerate import Accelerator
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 from utils_qa import postprocess_qa_predictions
@@ -381,9 +381,8 @@ def main():
             if repo_name is None:
                 repo_name = Path(args.output_dir).absolute().name
             # Create repo and retrieve repo_id
-            repo_id = create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
-            # Clone repo locally
-            repo = Repository(args.output_dir, clone_from=repo_id, token=args.hub_token)
+            api = HfApi()
+            repo_id = api.create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
 
             with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                 if "step_*" not in gitignore:
@@ -912,8 +911,12 @@ def main():
             )
             if accelerator.is_main_process:
                 tokenizer.save_pretrained(args.output_dir)
-                repo.push_to_hub(
-                    commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True
+                api.upload_folder(
+                    commit_message=f"Training in progress epoch {epoch}",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
                 )
 
     # Evaluation
@@ -1013,8 +1016,13 @@ def main():
         if accelerator.is_main_process:
             tokenizer.save_pretrained(args.output_dir)
             if args.push_to_hub:
-                repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
-
+                api.upload_folder(
+                    commit_message="End of training",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
+                )
             logger.info(json.dumps(eval_metric, indent=4))
             save_prefixed_metrics(eval_metric, args.output_dir)
 

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -29,7 +29,7 @@ from accelerate import Accelerator
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
-from huggingface_hub import Repository, create_repo, hf_hub_download
+from huggingface_hub import HfApi, hf_hub_download
 from PIL import Image
 from torch.utils.data import DataLoader
 from torchvision import transforms
@@ -365,9 +365,8 @@ def main():
             if repo_name is None:
                 repo_name = Path(args.output_dir).absolute().name
             # Create repo and retrieve repo_id
-            repo_id = create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
-            # Clone repo locally
-            repo = Repository(args.output_dir, clone_from=repo_id, token=args.hub_token)
+            api = HfApi()
+            repo_id = api.create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
 
             with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                 if "step_*" not in gitignore:
@@ -632,10 +631,12 @@ def main():
                         )
                         if accelerator.is_main_process:
                             image_processor.save_pretrained(args.output_dir)
-                            repo.push_to_hub(
-                                commit_message=f"Training in progress {completed_steps} steps",
-                                blocking=False,
-                                auto_lfs_prune=True,
+                            api.upload_folder(
+                                commit_message=f"Training in progress epoch {epoch}",
+                                folder_path=args.output_dir,
+                                repo_id=repo_id,
+                                repo_type="model",
+                                token=args.hub_token,
                             )
 
             if completed_steps >= args.max_train_steps:
@@ -687,8 +688,12 @@ def main():
             )
             if accelerator.is_main_process:
                 image_processor.save_pretrained(args.output_dir)
-                repo.push_to_hub(
-                    commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True
+                api.upload_folder(
+                    commit_message=f"Training in progress epoch {epoch}",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
                 )
 
         if args.checkpointing_steps == "epoch":
@@ -709,7 +714,13 @@ def main():
         if accelerator.is_main_process:
             image_processor.save_pretrained(args.output_dir)
             if args.push_to_hub:
-                repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
+                api.upload_folder(
+                    commit_message="End of training",
+                    folder_path=args.output_dir,
+                    repo_id=repo_id,
+                    repo_type="model",
+                    token=args.hub_token,
+                )
 
             all_results = {
                 f"eval_{k}": v.tolist() if isinstance(v, np.ndarray) else v for k, v in eval_metrics.items()

examples/pytorch/speech-pretraining/run_wav2vec2_pretraining_no_trainer.py

@@ -27,7 +27,7 @@ import torch
 from accelerate import Accelerator
 from accelerate.logging import get_logger
 from datasets import DatasetDict, concatenate_datasets, load_dataset
-from huggingface_hub import Repository, create_repo
+from huggingface_hub import HfApi
 from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm
 
@@ -423,9 +423,14 @@ def main():
             if repo_name is None:
                 repo_name = Path(args.output_dir).absolute().name
             # Create repo and retrieve repo_id
-            repo_id = create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
-            # Clone repo locally
-            repo = Repository(args.output_dir, clone_from=repo_id, token=args.hub_token)
+            api = HfApi()
+            repo_id = api.create_repo(repo_name, exist_ok=True, token=args.hub_token).repo_id
+
+            with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
+                if "step_*" not in gitignore:
+                    gitignore.write("step_*\n")
+                if "epoch_*" not in gitignore:
+                    gitignore.write("epoch_*\n")
         elif args.output_dir is not None:
             os.makedirs(args.output_dir, exist_ok=True)
     accelerator.wait_for_everyone()
@@ -719,10 +724,12 @@ def main():
                     )
 
                 if (args.push_to_hub and epoch < args.num_train_epochs - 1) and accelerator.is_main_process:
-                    repo.push_to_hub(
-                        commit_message=f"Training in progress step {completed_steps}",
-                        blocking=False,
-                        auto_lfs_prune=True,
+                    api.upload_folder(
+                        commit_message=f"Training in progress epoch {epoch}",
+                        folder_path=args.output_dir,
+                        repo_id=repo_id,
+                        repo_type="model",
+                        token=args.hub_token,
                     )
 
             # if completed steps > `args.max_train_steps` stop
@@ -772,7 +779,13 @@ def main():
             )
             if accelerator.is_main_process:
                 if args.push_to_hub:
-                    repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
+                    api.upload_folder(
+                        commit_message="End of training",
+                        folder_path=args.output_dir,
+                        repo_id=repo_id,
+                        repo_type="model",
+                        token=args.hub_token,
+                    )
 
 
 if __name__ == "__main__":