Internal Server Error when pushing local folder to repo branch

[lewtun]

Describe the bug

When trying to push a local Hub repo to another Hub repo's branch with upload_folder(), I am hitting an error like the following:

HfHubHTTPError: 500 Server Error: Internal Server Error for url: https://huggingface.co/api/models/lewtun/test-upload-folder/commit/test-branch?create_pr=1 (Request ID: Root=1-641989f5-38dfc6d84e63800f08d7fede)

Curiously, the error is resolved if I copy the local Hub repo to another directory and point upload_folder() to that. Concretely:

from huggingface_hub import Repository, upload_folder, create_branch, create_repo

# Create dummy repo we want to push files to
create_repo("test-upload-folder", repo_type="model")

# Create branch we want to push to
create_branch(repo_id="lewtun/test-upload-folder", branch="test-branch")

# Clone a Hub repo we want to push the contents to `test-upload-folder`
!mkdir -p dummy-model
repo = Repository(local_dir="dummy-model", clone_from="lewtun/dummy-trl-model")

# Push to branch - fails
upload_folder(repo_id="lewtun/test-upload-folder", folder_path="dummy-model", revision="test-branch", create_pr=True)

# Copy the local repo
!mkdir -p dummy-model-2
!cp -r dummy-model/* dummy-model-2

# Works?
upload_folder(repo_id="lewtun/test-upload-folder", folder_path="dummy-model-2", revision="test-branch", create_pr=True)

Perhaps there's some weird interaction with Git LFS at play here?

Reproduction

Run the cells in this Colab to reproduce the error: https://colab.research.google.com/drive/1lhiiw22EmIypYChZpbwVWwHcSNJpmqXh?usp=sharing

Logs

---------------------------------------------------------------------------
HTTPError                                 Traceback (most recent call last)
File /fsx/lewis/miniconda/envs/h4/lib/python3.10/site-packages/huggingface_hub/utils/_errors.py:259, in hf_raise_for_status(response, endpoint_name)
    258 try:
--> 259     response.raise_for_status()
    260 except HTTPError as e:

File /fsx/lewis/miniconda/envs/h4/lib/python3.10/site-packages/requests/models.py:1021, in Response.raise_for_status(self)
   1020 if http_error_msg:
-> 1021     raise HTTPError(http_error_msg, response=self)

HTTPError: 500 Server Error: Internal Server Error for url: https://huggingface.co/api/models/lewtun/distilgpt2-ift/commit/alpaca?create_pr=1

The above exception was the direct cause of the following exception:

HfHubHTTPError                            Traceback (most recent call last)
Cell In[54], line 1
----> 1 api.upload_folder(repo_id=TARGET_MODEL_ID, folder_path="distilgpt-ift/", revision="alpaca", create_pr=True)

File /fsx/lewis/miniconda/envs/h4/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py:120, in validate_hf_hub_args.<locals>._inner_fn(*args, **kwargs)
    117 if check_use_auth_token:
    118     kwargs = smoothly_deprecate_use_auth_token(fn_name=fn.__name__, has_token=has_token, kwargs=kwargs)
--> 120 return fn(*args, **kwargs)

File /fsx/lewis/miniconda/envs/h4/lib/python3.10/site-packages/huggingface_hub/hf_api.py:2790, in HfApi.upload_folder(self, repo_id, folder_path, path_in_repo, commit_message, commit_description, token, repo_type, revision, create_pr, parent_commit, allow_patterns, ignore_patterns, delete_patterns)
   2785     delete_operations = [
   2786         delete_op for delete_op in delete_operations if delete_op.path_in_repo not in added_paths
   2787     ]
   2788 commit_operations = delete_operations + add_operations
-> 2790 commit_info = self.create_commit(
   2791     repo_type=repo_type,
   2792     repo_id=repo_id,
   2793     operations=commit_operations,
   2794     commit_message=commit_message,
   2795     commit_description=commit_description,
   2796     token=token,
   2797     revision=revision,
   2798     create_pr=create_pr,
   2799     parent_commit=parent_commit,
   2800 )
   2802 if commit_info.pr_url is not None:
   2803     revision = quote(_parse_revision_from_pr_url(commit_info.pr_url), safe="")

File /fsx/lewis/miniconda/envs/h4/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py:120, in validate_hf_hub_args.<locals>._inner_fn(*args, **kwargs)
    117 if check_use_auth_token:
    118     kwargs = smoothly_deprecate_use_auth_token(fn_name=fn.__name__, has_token=has_token, kwargs=kwargs)
--> 120 return fn(*args, **kwargs)

File /fsx/lewis/miniconda/envs/h4/lib/python3.10/site-packages/huggingface_hub/hf_api.py:2443, in HfApi.create_commit(self, repo_id, operations, commit_message, commit_description, token, repo_type, revision, create_pr, num_threads, parent_commit)
   2441 try:
   2442     commit_resp = requests.post(url=commit_url, headers=headers, data=data, params=params)
-> 2443     hf_raise_for_status(commit_resp, endpoint_name="commit")
   2444 except RepositoryNotFoundError as e:
   2445     e.append_to_message(_CREATE_COMMIT_NO_REPO_ERROR_MESSAGE)

File /fsx/lewis/miniconda/envs/h4/lib/python3.10/site-packages/huggingface_hub/utils/_errors.py:301, in hf_raise_for_status(response, endpoint_name)
    297     raise BadRequestError(message, response=response) from e
    299 # Convert `HTTPError` into a `HfHubHTTPError` to display request information
    300 # as well (request id and/or server error message)
--> 301 raise HfHubHTTPError(str(e), response=response) from e

HfHubHTTPError: 500 Server Error: Internal Server Error for url: https://huggingface.co/api/models/lewtun/distilgpt2-ift/commit/alpaca?create_pr=1 (Request ID: Root=1-641987ae-38aa33eb3d1ef701354c2471)

Internal Error - We're working hard to fix that as soon as possible!

### System info

```shell
- huggingface_hub version: 0.13.3
- Platform: Linux-5.10.147+-x86_64-with-glibc2.31
- Python version: 3.9.16
- Running in iPython ?: No
- Running in notebook ?: No
- Running in Google Colab ?: No
- Token path ?: /root/.cache/huggingface/token
- Has saved token ?: True
- Who am I ?: lewtun
- Configured git credential helpers: store
- FastAI: 2.7.11
- Tensorflow: 2.11.0
- Torch: 1.13.1+cu116
- Jinja2: 3.1.2
- Graphviz: 0.20.1
- Pydot: 1.4.2
- Pillow: 8.4.0
- hf_transfer: N/A
- ENDPOINT: https://huggingface.co
- HUGGINGFACE_HUB_CACHE: /root/.cache/huggingface/hub
- HUGGINGFACE_ASSETS_CACHE: /root/.cache/huggingface/assets
- HF_TOKEN_PATH: /root/.cache/huggingface/token
- HF_HUB_OFFLINE: False
- HF_HUB_DISABLE_TELEMETRY: False
- HF_HUB_DISABLE_PROGRESS_BARS: None
- HF_HUB_DISABLE_SYMLINKS_WARNING: False
- HF_HUB_DISABLE_IMPLICIT_TOKEN: False
- HF_HUB_ENABLE_HF_TRANSFER: False

[Wauplin]

Thanks for pointing that out @lewtun. I'll have a look in a bit

[lewtun]

Ah, I wonder if the problem is that the local repo has its own Git history in .git and that conflicts with the target repo?

[Wauplin]

That's what I wondered as well but then why the cp -r dummy-model/* dummy-model-2 doesn't copy the .git folder? It should happen in both cases, right?

But yes, it's a good explanation.

[Wauplin]

(in general, I was also thinking of respecting the .gitignore file when doing a upload_folder)

[Wauplin]

@lewtun you were right, it is indeed an issue raised when pushing the .git folder. I created a proper issue to track that in huggingface_hub (#1402) along with the .gitignore stuff + an issue in moon-landing to fix it server-side (i.e. return a HTTP 400 with message). Thanks for noticing and reported it :)

FYI doing cp -r dummy-model/* dummy-model-2 doesn't copy the hidden files, that why copying the folder "solved" your problem. If you want to copy a folder with its hidden files, you must do cp -r dummy-model/. dummy-model-2.

---

As a side note for the script itself:

• it's more recommended to use from huggingface_hub import login). Internally it's the same but the later will work if the notebook is converted to a script. In this case, the token will be prompted in the terminal. Also you can provide the token manually (i.e. login(token="hf_***") which is not the case with notebook_login).
• creating a repo with create_repo outputs a RepoUrl object. It is a str with other properties, especially repo_id. That way you can do repo_id = create_repo("test-upload-folder").repo_id => repo_id is set to "lewtun/test-upload-folder" automatically. It's helpful if you want to share a notebook and don't want to care about the implicit username.