Datasets:

Tasks:
Other
Modalities:
Text
Formats:
parquet
Languages:
English
ArXiv:
Libraries:
Datasets
Dask
License:

Adds support for streaming. The code is based on the tfrecord loader from TorchData.

cc @albertvillanova @VictorSanh

Note for reviewers: The relevant changes reside in P3.py and io_utils.py. The rest are formatting changes made to be aligned with the datasets formatting style.

mariosasko changed pull request status to open
BigScience Workshop org

Awesome ! Just one nit: it seems that https://huggingface.co/datasets/bigscience/P3/blob/refs%2Fpr%2F11/tasks_splits_and_features.py is stored as an LFS file, which makes it not viewable in the UI

Regarding the dependencies - it seems that one just needs to pip install google to have this dataset work right ?

protobuf is the name of the dependency (pip install protobuf). I added a From: comment to the import to improve the error message when the lib is not installed. Since Tensorflow depends on it, the viewer should work out of the box.

BigScience Workshop org

Awesome ! LGTM then :)

mariosasko changed pull request status to merged

Sign up or log in to comment