kagglehub

• General Availability release

• Add output_dir parameter to dataset_download, model_download, notebook_output_download and competition_download for direct download to custom directory. (#283, #284, #285)

Release 0.4.2 (#282)

http://b/479835274

Release 0.4.1 (#280)

http://b/477283660

• Restored version-checking logic. (#272)
• Document new auth technique. (#270)
• Use kagglesdk instead of custom HTTP implementation. (#268)
• Dropped support for Python 3.9. The minimum required Python version is now 3.10. (#260)

Prepare release v0.3.13 (#257)

As titled.

This is to fix [Issue
#254](#254)

Prepare 0.3.12 release (#242)

http://b/388077145

[API] Update user agent for torchtune (#237)

Enable tracking of torchtune models via kagglehub api. This is the
change for kagglehub part. PR for midtier change is
http://go/kaggle-pr/34437

Fix = [b/381021004](https://buganizer.corp.google.com/issues/381021004)

Note: To test this change without midtier changes in production, you
need to have a midtier running locally and add breakpoint to review the
value for `userAgent` in `ProtoHttpRuleController.ApiV1.cs.` With that:

- Creating a python virtual environment

- Import this kagglehub version `pip install
git+https://github.com/Kaggle/kagglehub.git@nan_torch`

- Import torchtune and its required dependencies

- set environment variables
export KAGGLE_USERNAME=[your_user_name]
export KAGGLE_KEY=local_api_token

- in virtual environment, run command `tune download
"google/bert/tensorFlow2/answer-equivalence-bem" --source "kaggle"`

- verify user agent has torchtune and its associated version
<img width="650" alt="Screenshot 2025-03-30 at 12 18 49 PM"
src="https://github.com/user-attachments/assets/3993667f-489c-415a-9446-073bc926089c"
/>

• Fix Colab environment detection logic (#230)

• Renamed load_datset to dataset_load (#228)
• Support for Kaggle Packages (#196)

• Moved signing as optional feature due to dependency issue (#225)

• Add data loaders to user agent when downloading (#215)
• Sign model and publish to sigstore (#211)
• Use version number in notebook cache directories (#212)
• Notebook Output Download versioning (#206)
• Add support for utility scripts (#207)
• Fix load_dataset typo in README (#202)

• Add kagglehub.load_dataset(...) (#192, #197)
• Fix auto-compressed dataset downloads (#194, #195)
• Update docker-hatch and documentation (#193)

• Include server-side errors if present (#183)
• Made file logging opt-out by default (#190)
• Improved version upgrade warning (#188)
• Add kagglehub.notebook_output_download(...), only supports HTTP resolver for now. (#184)

• Add a naive Colab dataset cache resolver (#177)
• Add keras_hub as user-agent (#176)

• Hide API key in terminal when prompting in kaggle.login(...) (#173)
• Reduce logging spam when using force_download in Kaggle notebook environment (#172)

• Fallback to next auth mechanism if auth environment variables are set but empty (#168)
• kaggle.whoami() returns proper value in Kaggle Notebook environment (#166)
• kaggle.whoami() returns proper value when using Colab secrets (#165)

• Attach competition in Kaggle notebook environment (#161)
• Download competition using kagglehub.competition_download(...) (#158)
• Enable authorization via Colab secrets (#157)

• Integrate Keras metrics (#152)
• Enhance logging messages (#151)
• Upgrade Python version support to 3.9 and above (#154)

• Improve upload by ignoring patterns (#147)
• Upload with empty files (#145)
• Attach dataset in Kaggle notebook environment (#134)
• Upload datasets using kagglehub.dataset_upload(...) (#149)
• Download datasets using kagglehub.dataset_download(...) (#131)

• Added support for implicit token auth in kaggle notebooks (#141)
• Improve logging for cached models in Kaggle & Colab environments (#142)

• Fix model downloading bug

• Add support for .zip archive (#135)
• Reduce spammy logging (#124)
• Download model files in parallel (#122)

• Add missing packaging dependencies (#115)
• Create whoami method (#114)
• Migrate all tests to use Flask (#111)

• Create a new version when model instance exist but no versions are ready yet (#108)
• Remove zipping in upload (#105)

• Improve upload speed (#100)

• Add support for single file upload (#97)

• Add support for directory upload (#82)

• Add raise_for_status() in post function (#82)
• Use Artifact Registry for helper images (#83)

• Fix message when detecting newer version (#82)
• Link to model detail page in errors(Colab resolver) (#83)

• Include URL to model detail page in error message (#80)
• Add Kaggle/Colab to user-agent if running on these environment (#78)
• Improve logging for Colab resolver (#77)

• Fix model_upload with nested directory (#75)
• Detect if a newer version of kagglehub is available and suggest to upgrade (#73)

• Fix permission issue in model_upload and add integration tests (#69)
• Make specifying a license optional in model_upload (#62)
• Improve logging (#68, #71)
• Add resumable upload (#55)

• Prevent log message from being printed twice in some environment (#57)
• Add Colab model resolver (#53)
• Add kagglehub.model_upload(...) (#43, #51, #52)
• Add kagglehub user agent to Kaggle API V1 calls (#50)
• Add force_download option to kagglehub.model_download() (#44)

• Improve error messages for KaggleCacheResolver (#40)

• Improve error messages for Kaggle API calls (#38)
• Perform integrity check after file download (#37)

• Fixed notebook environment detection logic (#36)

• Fixed login credential validation (#33, #34)

• Attach model in Kaggle notebook environment with internet disabled (#27)
• Login via IPyWidgets in notebook (#28)
• Login via prompt in terminal (#23)
• Attach model in Kaggle Notebook environment (#19)
• Support resumable download (#17)
• Support unversioned model handles (#16)

• Login via environment variable or credentials file (#9)
• Download public models over HTTP and store in local cache (#8, #12)

• Skeleton for the kagglehub library (#1)