Dataset scripts are no longer supported, but found superb.py

[edwinzajac]

Describe the bug

Hello,

I'm trying to follow the Hugging Face Pipelines tutorial but the tutorial seems to work only on old datasets versions.

I then get the error :

--------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[65], [line 1](vscode-notebook-cell:?execution_count=65&line=1)
----> [1](vscode-notebook-cell:?execution_count=65&line=1) dataset = datasets.load_dataset("superb", name="asr", split="test")
      3 # KeyDataset (only *pt*) will simply return the item in the dict returned by the dataset item
      4 # as we're not interested in the *target* part of the dataset. For sentence pair use KeyPairDataset
      5 for out in tqdm(pipe(KeyDataset(dataset, "file"))):

File ~/Desktop/debug/llm_course/.venv/lib/python3.11/site-packages/datasets/load.py:1392, in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, verification_mode, keep_in_memory, save_infos, revision, token, streaming, num_proc, storage_options, **config_kwargs)
   1387 verification_mode = VerificationMode(
   1388     (verification_mode or VerificationMode.BASIC_CHECKS) if not save_infos else VerificationMode.ALL_CHECKS
   1389 )
   1391 # Create a dataset builder
-> [1392](https://file+.vscode-resource.vscode-cdn.net/home/edwin/Desktop/debug/llm_course/~/Desktop/debug/llm_course/.venv/lib/python3.11/site-packages/datasets/load.py:1392) builder_instance = load_dataset_builder(
   1393     path=path,
   1394     name=name,
   1395     data_dir=data_dir,
   1396     data_files=data_files,
   1397     cache_dir=cache_dir,
   1398     features=features,
   1399     download_config=download_config,
   1400     download_mode=download_mode,
   1401     revision=revision,
   1402     token=token,
   1403     storage_options=storage_options,
   1404     **config_kwargs,
   1405 )
   1407 # Return iterable dataset in case of streaming
   1408 if streaming:

File ~/Desktop/debug/llm_course/.venv/lib/python3.11/site-packages/datasets/load.py:1132, in load_dataset_builder(path, name, data_dir, data_files, cache_dir, features, download_config, download_mode, revision, token, storage_options, **config_kwargs)
   1130 if features is not None:
   1131     features = _fix_for_backward_compatible_features(features)
-> [1132](https://file+.vscode-resource.vscode-cdn.net/home/edwin/Desktop/debug/llm_course/~/Desktop/debug/llm_course/.venv/lib/python3.11/site-packages/datasets/load.py:1132) dataset_module = dataset_module_factory(
   1133     path,
   1134     revision=revision,
   1135     download_config=download_config,
   1136     download_mode=download_mode,
   1137     data_dir=data_dir,
   1138     data_files=data_files,
   1139     cache_dir=cache_dir,
   1140 )
   1141 # Get dataset builder class
   1142 builder_kwargs = dataset_module.builder_kwargs

File ~/Desktop/debug/llm_course/.venv/lib/python3.11/site-packages/datasets/load.py:1031, in dataset_module_factory(path, revision, download_config, download_mode, data_dir, data_files, cache_dir, **download_kwargs)
   1026             if isinstance(e1, FileNotFoundError):
   1027                 raise FileNotFoundError(
   1028                     f"Couldn't find any data file at {relative_to_absolute_path(path)}. "
   1029                     f"Couldn't find '{path}' on the Hugging Face Hub either: {type(e1).__name__}: {e1}"
   1030                 ) from None
-> [1031](https://file+.vscode-resource.vscode-cdn.net/home/edwin/Desktop/debug/llm_course/~/Desktop/debug/llm_course/.venv/lib/python3.11/site-packages/datasets/load.py:1031)             raise e1 from None
   1032 else:
   1033     raise FileNotFoundError(f"Couldn't find any data file at {relative_to_absolute_path(path)}.")

File ~/Desktop/debug/llm_course/.venv/lib/python3.11/site-packages/datasets/load.py:989, in dataset_module_factory(path, revision, download_config, download_mode, data_dir, data_files, cache_dir, **download_kwargs)
    981 try:
    982     api.hf_hub_download(
    983         repo_id=path,
    984         filename=filename,
   (...)    987         proxies=download_config.proxies,
    988     )
--> [989](https://file+.vscode-resource.vscode-cdn.net/home/edwin/Desktop/debug/llm_course/~/Desktop/debug/llm_course/.venv/lib/python3.11/site-packages/datasets/load.py:989)     raise RuntimeError(f"Dataset scripts are no longer supported, but found {filename}")
    990 except EntryNotFoundError:
    991     # Use the infos from the parquet export except in some cases:
    992     if data_dir or data_files or (revision and revision != "main"):

RuntimeError: Dataset scripts are no longer supported, but found superb.py

NB : I tried to replace "superb" by "anton-l/superb_demo" but I get a 'torchcodec' importing error. Maybe I misunderstood something.

Steps to reproduce the bug

import datasets
from transformers import pipeline
from transformers.pipelines.pt_utils import KeyDataset
from tqdm.auto import tqdm

pipe = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h", device=0)
dataset = datasets.load_dataset("superb", name="asr", split="test")

# KeyDataset (only *pt*) will simply return the item in the dict returned by the dataset item
# as we're not interested in the *target* part of the dataset. For sentence pair use KeyPairDataset
for out in tqdm(pipe(KeyDataset(dataset, "file"))):
    print(out)
    # {"text": "NUMBER TEN FRESH NELLY IS WAITING ON YOU GOOD NIGHT HUSBAND"}
    # {"text": ....}
    # ....

Expected behavior

Get the tutorial expected results

Environment info

--- SYSTEM INFO ---
Operating System: Ubuntu 24.10
Kernel: Linux 6.11.0-29-generic
Architecture: x86-64

--- PYTHON ---
Python 3.11.13

--- VENV INFO ----
datasets=4.0.0
transformers=4.53
tqdm=4.67.1

[dejokz]

I got a pretty similar issue when I try to load bigbio/neurotrial_ner dataset.
Dataset scripts are no longer supported, but found neurotrial_ner.py

[gMontoyaSpeech]

Same here. I was running this tutorial and got a similar error: openai/whisper#654 (I'm a first-time transformers library user)

RuntimeError: Dataset scripts are no longer supported, but found librispeech_asr.py

What am I supposed to do at this point?

Thanks

[Tin-viAct]

hey I got the same error and I have tried to downgrade version to 3.6.0 and it works.
pip install datasets==3.6.0

[gMontoyaSpeech]

Thank you very much @Tin-viAct . That indeed did the trick for me :)
Now the code continue its normal flow

[johnbarb71]

Thanks @Tin-viAct, Works!

[lhoestq]

I converted openslr/librispeech_asr to Parquet - thanks for reporting.

It's now compatible with datasets 4.0 !

I'll try to ping the authors of the other datasets like s3prl/superb and espnet/yodas2

[pgzmnk]

How come a breaking change was allowed and now requires extra work from individual authors for things to be usable?

https://en.wikipedia.org/wiki/Backward_compatibility

[lhoestq]

We follow semantic versioning so that breaking changes only occur in major releases. Also note that dataset scripts have been legacy for some time now, with a message on the dataset pages to ask authors to update their datasets.

It's ok to ping older versions of datasets, but imo a few remaining datasets need to be converted since they are valuable to the community.