Add Huggingface Integration #916
Conversation
|
MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅ |
There was a problem hiding this comment.
Before I start reviewing this in earnest, I would need at least the following 2 pieces of information to be added to the PR:
- Documentation: (it is absolutely fine to have a bullet point list of items that link to the main HF docs)
- Tests
I believe both of these were present in the previous PR.
|
Hi @Wauplin and @NielsRogge - this PR looks good from my end. Do you have any feedback? |
There was a problem hiding this comment.
Hey there 👋 Thanks for the ping!
I left a few comments from an outsider point of view. I do think that the CLI should be more opinionated (understand "have less options and decide things for the user") otherwise we pretty much end up with a CLI close to what huggingface-cli upload and huggingface-cli download do.
| from pathlib import Path | ||
| from GANDLF.utils import get_git_hash | ||
|
|
||
| readme_template = """ |
There was a problem hiding this comment.
Is this a simple copy of the model card template found here? If yes, I can suggest to either:
- directly reuse the template from
huggingface_hub(i.e.ModelCard.from_template(card_data)without thetemplate_str). - or define your own template but in this case you should only put the relevant fields and descriptions for your library (instead of having all fields as empty)
There was a problem hiding this comment.
Thanks @Wauplin for making me aware of this ,I will definitely go through it and make required changes as you mentioned
There was a problem hiding this comment.
Hi @Wauplin,
We had an internal discussion on what would be the best way for us to showcase potential model uploaders with a specific set of required options for the model card. Thus far, we have landed on using a custom model card. The reason to have all the fields present is provide the ability for a user to put in more information than what we require.
Here, we have put the string "REQUIRED_FOR_GANDLF" for the fields that are explicitly needed for the user to populate, and the rest have been left as present in the template.
In the code, we plan to add 2 checks:
- If
"REQUIRED_FOR_GANDLF"is found, we present an error to the user saying that this field needs to be populated with appropriate information. - The
Repositorykey should always behttps://github.com/mlcommons/GaNDLF.
Thoughts?
There was a problem hiding this comment.
Brilliant, thanks for the confirmation! We'll get on it right away. 😄
There was a problem hiding this comment.
@sarthakpati @Wauplin so how can we test this file if we propose the upload functionality as we only have entry points tests, do we have to mention a specific directory there
There was a problem hiding this comment.
Perhaps you can leverage one of the existing training tests to test the upload. I would recommend this one, since this would only upload a single model.
Ensure you put an appropriate description for it (such as Unit testing model or something) to make it clear for anyone viewing it. Is there a way to update an existing model, @Wauplin?
| tags += [git_hash] | ||
|
|
||
| card_data = ModelCardData(library_name="GaNDLF", tags=tags) | ||
| card = ModelCard.from_template(card_data, template_str=readme_template) |
| def download_from_hub( | ||
| repo_id: str, | ||
| revision: Union[str, None] = None, | ||
| cache_dir: Union[str, None] = None, | ||
| local_dir: Union[str, None] = None, | ||
| force_download: bool = False, | ||
| token: Union[str, None] = None, | ||
| ): | ||
| snapshot_download( | ||
| repo_id=repo_id, | ||
| revision=revision, | ||
| cache_dir=cache_dir, | ||
| local_dir=local_dir, | ||
| force_download=force_download, | ||
| token=token, | ||
| ) |
There was a problem hiding this comment.
I am not sure this alias is really needed. I would simply call snapshot_download in other places in the code.
There was a problem hiding this comment.
I still think the alias is not needed and that snapshot_download could be used by default
Co-authored-by: Lucain <lucainp@gmail.com>
|
@pranayasinghcsmpl some lint fixes (unused variables and whatnot) will be needed for this PR. Thanks for taking care of it! Thanks for your comments and suggestions, @Wauplin! |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #916 +/- ##
==========================================
+ Coverage 94.58% 94.61% +0.03%
==========================================
Files 161 164 +3
Lines 9567 9701 +134
==========================================
+ Hits 9049 9179 +130
- Misses 518 522 +4
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
|
Support ticket generated with Codacy to explore the coverage issue. |
|
Codacy folks suggested not to use coverage reporter for anything coming in from other forks 🙄 Anyway, we should be good to go from my end. @Wauplin is this PR good to merge for you? |
| api = HfApi(token=token) | ||
|
|
||
| try: | ||
| api.create_repo(repo_id) | ||
| except Exception as e: | ||
| print(f"Error: {e}") | ||
|
|
||
| api = HfApi(token=token) | ||
|
|
||
| repo_id = api.create_repo(repo_id, exist_ok=True).repo_id |
There was a problem hiding this comment.
| api = HfApi(token=token) | |
| try: | |
| api.create_repo(repo_id) | |
| except Exception as e: | |
| print(f"Error: {e}") | |
| api = HfApi(token=token) | |
| repo_id = api.create_repo(repo_id, exist_ok=True).repo_id | |
| api = HfApi(token=token) | |
| try: | |
| repo_id = api.create_repo(repo_id).repo_id | |
| except Exception as e: | |
| print(f"Error: {e}") |
no need to create the repo twice
|
|
||
| api.upload_folder( | ||
| repo_id=repo_id, | ||
| token=token, |
There was a problem hiding this comment.
| token=token, |
no need for this since already provided in HfApi
| def download_from_hub( | ||
| repo_id: str, | ||
| revision: Union[str, None] = None, | ||
| cache_dir: Union[str, None] = None, | ||
| local_dir: Union[str, None] = None, | ||
| force_download: bool = False, | ||
| token: Union[str, None] = None, | ||
| ): | ||
| snapshot_download( | ||
| repo_id=repo_id, | ||
| revision=revision, | ||
| cache_dir=cache_dir, | ||
| local_dir=local_dir, | ||
| force_download=force_download, | ||
| token=token, | ||
| ) |
There was a problem hiding this comment.
I still think the alias is not needed and that snapshot_download could be used by default
| @click.option( | ||
| "--allow-patterns", | ||
| "-ap", | ||
| help="Uploading: If provided, only files matching at least one pattern are uploaded.", | ||
| ) | ||
| @click.option( | ||
| "--ignore-patterns", | ||
| "-ip", | ||
| help="Uploading: If provided, files matching any of the patterns are not uploaded.", | ||
| ) | ||
| @click.option( | ||
| "--delete-patterns", | ||
| "-dp", | ||
| help="Uploading: If provided, remote files matching any of the patterns will be deleted from the repo while committing new files. This is useful if you don't know which files have already been uploaded.", | ||
| ) |
There was a problem hiding this comment.
| @click.option( | |
| "--allow-patterns", | |
| "-ap", | |
| help="Uploading: If provided, only files matching at least one pattern are uploaded.", | |
| ) | |
| @click.option( | |
| "--ignore-patterns", | |
| "-ip", | |
| help="Uploading: If provided, files matching any of the patterns are not uploaded.", | |
| ) | |
| @click.option( | |
| "--delete-patterns", | |
| "-dp", | |
| help="Uploading: If provided, remote files matching any of the patterns will be deleted from the repo while committing new files. This is useful if you don't know which files have already been uploaded.", | |
| ) |
I don't think this is needed since you are in control of what needs to be uploaded, right?
| @click.option( | ||
| "--hf-template", | ||
| "-hft", | ||
| help="Adding the template path for the model card it is Required during Uploaing a model", | ||
| type=click.Path(exists=True, file_okay=True, dir_okay=False), | ||
| ) |
There was a problem hiding this comment.
Maybe default it to hugging_face.md to reduce friction? Users are free to provide another template if they want but having one by default should reduce friction and help grow usage.
| @@ -82,6 +82,7 @@ | |||
| "typer==0.9.0", | |||
| "colorlog", | |||
| "opacus==1.5.2", | |||
| "huggingface-hub==0.23.4", | |||
There was a problem hiding this comment.
(nit) latest is 0.25.1
| "huggingface-hub==0.23.4", | |
| "huggingface-hub==0.25.1", |
Fixes #727
Proposed Changes
Checklist
CONTRIBUTINGguide has been followed.typingis used to provide type hints, including and not limited to usingOptionalif a variable has a pre-defined value).pip installstep is needed for PR to be functional), please ensure it is reflected in all the files that control the CI, namely: python-test.yml, and all docker files [1,2,3].