`huggingface_hub.errors.HfHubHTTPError: 404 Client Error: Not Found for url` when fetching a dataset with `datasets.load_dataset`

[harupy]

Not a bug but a question. We started getting the following error:

https://github.com/mlflow/mlflow/actions/runs/21368603305/job/61506951617

ests/data/test_huggingface_dataset_and_source.py::test_from_huggingface_dataset_constructs_expected_dataset_with_revision - huggingface_hub.errors.HfHubHTTPError: 404 Client Error: Not Found for url: https://huggingface.co/api/datasets/rotten_tomatoes/revision/aa13bc287fa6fcab6daf52f0dfb9994269ffea28 (Request ID: Root=1-6977aeca-35bc2b5b605884926a9224d0;aa2391f3-26e8-4975-b9bb-114b2fa40223)

Adding a user id fixed the issue (mlflow/mlflow#20350). https://huggingface.co/api/datasets no longer accepts a name-only path like rotten_tomatoes? Just wondering what changed. Thanks!

[lhoestq]

Hi ! Yes you should use cornell-movie-review-data/rotten_tomatoes instead of rotten_tomatoes, which is the legacy name. Those datasets have been moved under their actual owners accounts some time ago (but we were keeping the old names as aliases)

Some other impacted names are:

• imdb -> stanfordnlp/imdb
• wikitext -> Salesforce/wikitext
• gsm8k -> openai/gsm8k
• winogrande -> allenai/winogrande

We're working on re-enabling them as aliases for backward compatibility. I'll post updates here, sorry for the inconvenience.

Using the actual name instead of the old legacy name is more future proof though

[kurt-essential]

Thanks for the heads up @lhoestq ! fyi, this change is likely breaking a lot of repos that have legacy names hardcoded (example) Would be helpful to many to share this update in a more visible way if it is likely to persist for a while.

[julien-c]

internal tracking link

[harupy]

@lhoestq Thanks for clarifying!

[lhoestq]

The aliases are re-enabled :)

[harupy]

Thanks!

[harupy]

Can I close this issue?