NotImplementedError when solidifying a streaming dataset


I’m trying to take only the top 100 rows from a very large dataset, using the following method from the documentation here.

dataset = load_dataset("laion/laion-coco", split="train", streaming=True)

However, this raises the error

NotImplementedError: os.walk is not extended to support URLs in streaming mode

The code from the documentation worked fine, I’ve only changed the dataset URL below.

Am I missing something? Any help would be great, thanks in advance.


  • MacOS Ventura 13.1
  • Python 3.10.6
  • Datasets 2.8.0

Also tried on Ubuntu 20.04.4 LTS with the same python and dataset versions.

The full stacktrace is here:

NotImplementedError                       Traceback (most recent call last)
Cell In[29], line 2
      1 dataset = load_dataset("laion/laion-coco", split="train", streaming=True)
----> 2 list(dataset.take(100))

File /opt/homebrew/Caskroom/miniconda/base/envs/fyp/lib/python3.10/site-packages/datasets/, in IterableDataset.__iter__(self)
    843         yield from self._iter_pytorch(worker_info)
    844         return
--> 846 for key, example in self._iter():
    847     if self.features:
    848         # `IterableDataset` automatically fills missing columns with None.
    849         # This is done with `_apply_feature_types_on_example`.
    850         yield _apply_feature_types_on_example(
    851             example, self.features, token_per_repo_id=self._token_per_repo_id
    852         )

File /opt/homebrew/Caskroom/miniconda/base/envs/fyp/lib/python3.10/site-packages/datasets/, in IterableDataset._iter(self)
    786 else:
    787     ex_iterable = self._ex_iterable
--> 788 yield from ex_iterable

File /opt/homebrew/Caskroom/miniconda/base/envs/fyp/lib/python3.10/site-packages/datasets/, in TakeExamplesIterable.__iter__(self)
    627 def __iter__(self):
--> 628     yield from islice(self.ex_iterable, self.n)

File /opt/homebrew/Caskroom/miniconda/base/envs/fyp/lib/python3.10/site-packages/datasets/, in ExamplesIterable.__iter__(self)
    112 def __iter__(self):
--> 113     yield from self.generate_examples_fn(**self.kwargs)

File /opt/homebrew/Caskroom/miniconda/base/envs/fyp/lib/python3.10/site-packages/datasets/, in _generate_examples_from_tables_wrapper.<locals>.wrapper(**kwargs)
    711 def wrapper(**kwargs):
    712     python_formatter = PythonFormatter()
--> 713     for key, table in generate_tables_fn(**kwargs):
    714         batch = python_formatter.format_batch(table)
    715         for i, example in enumerate(_batch_to_examples(batch)):

File /opt/homebrew/Caskroom/miniconda/base/envs/fyp/lib/python3.10/site-packages/datasets/packaged_modules/parquet/, in Parquet._generate_tables(self, files)
     61     if sorted( for field in schema) != sorted(self.config.columns):
     62         raise ValueError(
     63             f"Tried to load parquet data with columns '{self.config.columns}' with mismatching features '{self.config.features}'"
     64         )
---> 65 for file_idx, file in enumerate(itertools.chain.from_iterable(files)):
     66     with open(file, "rb") as f:
     67         parquet_file = pq.ParquetFile(f)

File /opt/homebrew/Caskroom/miniconda/base/envs/fyp/lib/python3.10/site-packages/datasets/download/, in _IterableFromGenerator.__iter__(self)
    791 def __iter__(self):
--> 792     yield from self.generator(*self.args, **self.kwargs)

File /opt/homebrew/Caskroom/miniconda/base/envs/fyp/lib/python3.10/site-packages/datasets/download/, in FilesIterable._iter_from_urlpaths(cls, urlpaths, use_auth_token)
    845     yield urlpath
    846 else:
--> 847     for dirpath, dirnames, filenames in xwalk(urlpath, use_auth_token=use_auth_token):
    848         # skipping hidden directories; prune the search
    849         # [:] for the in-place list modification required by os.walk
    850         # (only works for local paths as fsspec's walk doesn't support the in-place modification)
    851         dirnames[:] = sorted([dirname for dirname in dirnames if not dirname.startswith((".", "__"))])
    852         if xbasename(dirpath).startswith((".", "__")):
    853             # skipping hidden directories

File /opt/homebrew/Caskroom/miniconda/base/envs/fyp/lib/python3.10/site-packages/datasets/download/, in xwalk(urlpath, use_auth_token, **kwargs)
    558 else:
    559     # walking inside a zip in a private repo requires authentication
    560     if not rest_hops and (main_hop.startswith("http://") or main_hop.startswith("https://")):
--> 561         raise NotImplementedError("os.walk is not extended to support URLs in streaming mode")
    562     elif rest_hops and (rest_hops[0].startswith("http://") or rest_hops[0].startswith("https://")):
    563         url = rest_hops[0]

NotImplementedError: os.walk is not extended to support URLs in streaming mode
1 Like

Hi ! Sorry for the inconvenience, we are fixing this right now for datasets hosted on - will keep you posted

1 Like

It’s fixed now ! you can now stream data again from the HF Hub :slightly_smiling_face:

1 Like

I’m also facing this issue but with “bigcode/the-stack-dedup” dataset. The error message is “os.walk is not extended to support URLs in streaming mode” and I’m using datasets==2.9.0.

Hi ! Can you try again ?

Hi, it still shows the same error … Is clearing cache, upgrade datasets library required ?

Does this happen right away when you start iterating on the dataset ? Can you try updating huggingface-hub and make sure you have a working connection to from your python environment ?

This happen in the middle of the iterating process … I’ve updated huggingface-hub but the error is still there … The c4 dataset can be load in the same way without error … so I guess it is something specific to the-stack-dedup

Ok. Not sure what’s causing this exactly. I see two possibilities

  • either one of the file is a ZIP file, and therefore it tries to walk inside it
  • OR there is an issue with xisfile() not returning True for one of the parquet files

In any case we need to know which file is causing this. It should be possible by mocking to add some logging every time it’s called until the error raises. Its first argument is the URL of the file we’re looking for.

In Python, NotImplementedError is an exception that is raised when a method or function is not implemented, or when a subclass does not implement an abstract method from its parent class. This exception can be raised intentionally to signal that the functionality is not yet implemented, or it can be raised accidentally due to a coding mistake.

Check if the method or function is intentionally not implemented: If the NotImplementedError is raised intentionally, you can leave the method or function as is, or provide a placeholder implementation that raises the NotImplementedError. However, if the NotImplementedError is raised accidentally, you need to provide a proper implementation.

If you are dealing with an python abstract method, you can either provide a concrete implementation for the method in the subclass, or mark the subclass as abstract and force its children to implement the method.

I have the same question, is there any update?

Is this happening at the beginning if streaming or while streaming after some time ?