Hugging Face's logo

Datasets:
allenai
/
c4

Tasks:
Text Generation
Fill-Mask
Modalities:
Text
Sub-tasks:
language-modeling
masked-language-modeling
Languages:
Afrikaans
Amharic
Arabic
Size:
10B - 100B
ArXiv:
License:
Dataset card Viewer Files Community
15

Cannot load dataset from hf hub

#7
by at676 - opened
Discussion
at676
•
edited Jan 10

My code gets stuck when running the following: 

    from datasets import load_dataset
    traindata = load_dataset(
        'allenai/c4', 'allenai--c4', data_files={'train': 'en/c4-train.00000-of-01024.json.gz'}, split='train'
    )
    valdata = load_dataset(
        'allenai/c4', 'allenai--c4', data_files={'validation': 'en/c4-validation.00000-of-00008.json.gz'}, split='validation'
    )

I am seeing a lot of "Resolving dataset files" lines and then nothing happens. This started happening earlier today and multiple people have been able to reproduce it on different machines, so it is probably a HF side issue.

Downloading readme: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 41.1k/41.1k [00:00<00:00, 6.04MB/s]
Resolving data files: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1024/1024 [00:07<00:00, 138.61it/s]
Resolving data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1024/1024 [00:00<00:00, 462969.42it/s]
Resolving data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 7168/7168 [00:00<00:00, 529188.23it/s]
Resolving data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [00:00<00:00, 476794.77it/s]
Resolving data files: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 512/512 [00:00<00:00, 424403.88it/s]
lhoestq
•
edited Jan 10

Hi ! The "allenai--c4" configuration doesn't exist for this dataset (it's a legacy scheme from old versions of the datasets library)

You can try this instead:

from datasets import load_dataset
traindata = load_dataset(
    'allenai/c4', data_files={'train': 'en/c4-train.00000-of-01024.json.gz'}, split='train'
)
valdata = load_dataset(
    'allenai/c4', data_files={'validation': 'en/c4-validation.00000-of-00008.json.gz'}, split='validation'
)

Also make sure to use the latest versions of datasets and huggingface_hub

pip install -U datasets huggingface_hub
lhoestq

We recently added some examples on how to use the dataset in the dataset page btw: https://huggingface.co/datasets/allenai/c4

at676

Thanks, this does fix the problem. I had to upgrade from datasets 2.14.6 to 2.16.1, but 2.14.6 doesn't seem very old to me. Do you know what change on HF's servers caused this problem? How can I be notified of these changes in the future so I don't have to open a ticket and can proactively upgrade my code?

lhoestq

Hi ! This change happened in the allenai/c4 repo itself. We recently added a list of configurations users can select to load any variant of c4 (e.g. "en", "multilingual", etc.)

You can get notifications when this dataset is updated by Watching this repo (or watching all of allenai repositories for example). Note that dataset updates are pretty rare though.

Btw it's also possible to pin the git revision of the dataset you load with load_dataset and ignore future update

lemonflourorange

This change broke some tools which loaded c4 using an older version of datasets fyi

lhoestq

Yes unfortunately :/ I'm happy to open PRs to fix the impacted repos on github though, if you have some links to share

Note that it's still possible to use this repository before the change by pinning the old revision: revision="607bd4c8450a42878aa9ddc051a65a055450ef87"

dingyunan

ValueError: There are multiple 'allenai/c4' configurations in the cache: default-b04fc8a0b8562884, default-c7bc8b0aefc5e48f
Please specify which configuration to reload from the cache, e.g.

my code

cache_dir = '/mini/others/allenai'
traindata = load_dataset('allenai/c4', data_files={'train': 'en/c4-train.00000-of-01024.json.gz'}, split='train', cache_dir=cache_dir)
valdata = load_dataset('allenai/c4',  data_files={'validation': 'en/c4-validation.00000-of-00008.json.gz'}, split='validation',  cache_dir=cache_dir)
lhoestq

Which version of datasets are you using ? Using 3.0.1 it works fine on my side

Sign up or log in to comment