Datasets:

brianhuster

/

github-issues

like 0

https://api.github.com/repos/huggingface/datasets/issues/6759

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6759/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6759/comments

https://api.github.com/repos/huggingface/datasets/issues/6759/events

https://github.com/huggingface/datasets/issues/6759

I_kwDODunzps6DqQfb

Persistent multi-process Pool

open

NONE

### Feature request Running .map and filter functions with `num_procs` consecutively instantiates several multiprocessing pools iteratively. As instantiating a Pool is very resource intensive it can be a bottleneck to performing iteratively filtering. My ideas: 1. There should be an option to declare `persistent_workers` similar to pytorch DataLoader. Downside would be that would be complex to determine the correct resource allocation and deallocation of the pool. i.e. the dataset can outlive the utility of the pool. 2. Provide a pool as an argument. Downside would be the expertise required by the user. Upside, is that there is better resource management. ### Motivation Is really slow to iteratively perform map and filter operations on a dataset. ### Your contribution If approved I could integrate it. I would need to know what method would be most suitable to implement from the two options above.

https://api.github.com/repos/huggingface/datasets/issues/6759/timeline

https://api.github.com/repos/huggingface/datasets/issues/6758

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6758/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6758/comments

https://api.github.com/repos/huggingface/datasets/issues/6758/events

https://github.com/huggingface/datasets/issues/6758

I_kwDODunzps6DovLe

Passing `sample_by` to `load_dataset` when loading text data does not work

open

NONE

### Describe the bug I have a dataset that consists of a bunch of text files, each representing an example. There is an undocumented `sample_by` argument for the `TextConfig` class that is used by `Text` to decide whether to split files into lines, paragraphs or take them whole. Passing `sample_by=“document”` to `load_dataset` results in files getting split into lines regardless. I have edited `src/datasets/packaged_modules/text/text.py` for myself to switch the default and it works fine. As a side note, the `if-else` for `sample_by` will silently load an empty dataset if someone makes a typo in the argument, which is not ideal. ### Steps to reproduce the bug 1. Prepare data as a bunch of files in a directory. 2. Load that data via `load_dataset(“text”, data_files=<data_dir>/<files_glob>, …, sample_by=“document”)`. 3. Inspect the resultant dataset — every item should have the form of `{“text”: <a line from a file>}`. ### Expected behavior `load_dataset(“text”, data_files=<data_dir>/<files_glob>, …, sample_by=“document”)` should result in a dataset with items of the form `{“text”: <one document>}`. ### Environment info - `datasets` version: 2.18.0 - Platform: Linux-5.15.0-1046-nvidia-x86_64-with-glibc2.35 - Python version: 3.11.8 - `huggingface_hub` version: 0.21.4 - PyArrow version: 15.0.2 - Pandas version: 2.2.1 - `fsspec` version: 2024.2.0

https://api.github.com/repos/huggingface/datasets/issues/6758/timeline

https://api.github.com/repos/huggingface/datasets/issues/6757

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6757/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6757/comments

https://api.github.com/repos/huggingface/datasets/issues/6757/events

https://github.com/huggingface/datasets/pull/6757

PR_kwDODunzps5qr7Li

Test disabling transformers containers in docs CI

open

CONTRIBUTOR

Related to https://github.com/huggingface/doc-builder/pull/487 and [internal slack thread](https://huggingface.slack.com/archives/C04F8N7FQNL/p1711384899462349?thread_ts=1711041424.720769&cid=C04F8N7FQNL). There is now a `custom_container` option when building docs in CI. When set to `""` (instead of `"huggingface/transformers-doc-builder"` by default), we don't run the CI inside a container, therefore saving ~2min of download time. The plan is to test disabling the transformers container on a few "big" repo and if everything works correctly, we will stop making it the default container. More details on https://github.com/huggingface/doc-builder/pull/487. cc @mishig25

https://api.github.com/repos/huggingface/datasets/issues/6757/timeline

https://api.github.com/repos/huggingface/datasets/issues/6756

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6756/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6756/comments

https://api.github.com/repos/huggingface/datasets/issues/6756/events

https://github.com/huggingface/datasets/issues/6756

I_kwDODunzps6DdiPd

Support SQLite files?

closed

CONTRIBUTOR

### Feature request Support loading a dataset from a SQLite file https://huggingface.co/datasets/severo/test_iris_sqlite/tree/main ### Motivation SQLite is a popular file format. ### Your contribution See discussion on slack: https://huggingface.slack.com/archives/C04L6P8KNQ5/p1702481859117909 (internal) In particular: a SQLite file can contain multiple tables, which might be matched to multiple configs. Maybe the detail of splits and configs should be defined in the README YAML, or use the same format as for ZIP files: `Iris.sqlite::Iris`. See dataset here: https://huggingface.co/datasets/severo/test_iris_sqlite Note: should we also support DuckDB files?

https://api.github.com/repos/huggingface/datasets/issues/6756/timeline

completed

https://api.github.com/repos/huggingface/datasets/issues/6755

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6755/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6755/comments

https://api.github.com/repos/huggingface/datasets/issues/6755/events

https://github.com/huggingface/datasets/issues/6755

I_kwDODunzps6DZx5p

Small typo on the documentation

open

NONE

### Describe the bug There is a small typo on https://github.com/huggingface/datasets/blob/d5468836fe94e8be1ae093397dd43d4a2503b926/src/datasets/dataset_dict.py#L938 It should be `caching is enabled`. ### Steps to reproduce the bug Please visit https://github.com/huggingface/datasets/blob/d5468836fe94e8be1ae093397dd43d4a2503b926/src/datasets/dataset_dict.py#L938 ### Expected behavior `caching is enabled` ### Environment info - `datasets` version: 2.17.1 - Platform: Linux-5.15.0-101-generic-x86_64-with-glibc2.35 - Python version: 3.11.7 - `huggingface_hub` version: 0.20.3 - PyArrow version: 15.0.0 - Pandas version: 2.2.1 - `fsspec` version: 2023.10.0

https://api.github.com/repos/huggingface/datasets/issues/6755/timeline

https://api.github.com/repos/huggingface/datasets/issues/6754

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6754/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6754/comments

https://api.github.com/repos/huggingface/datasets/issues/6754/events

https://github.com/huggingface/datasets/pull/6754

PR_kwDODunzps5qk-nr

Fix cache path to snakecase for `CachedDatasetModuleFactory` and `Cache`

open

NONE

Fix https://github.com/huggingface/datasets/issues/6750#issuecomment-2016678729 I didn't find a guideline on how to run the tests, so i just run the following steps to make sure that this bug is fixed. 1. `python test.py`, 2. then `HF_DATASETS_OFFLINE=1 python test.py` The `test.py` is ``` import datasets datasets.utils.logging.set_verbosity_info() ds = datasets.load_dataset('izhx/STS17-debug') print(ds) ds = datasets.load_dataset('C-MTEB/AFQMC', revision='b44c3b011063adb25877c13823db83bb193913c4') print(ds) ```

https://api.github.com/repos/huggingface/datasets/issues/6754/timeline

https://api.github.com/repos/huggingface/datasets/issues/6753

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6753/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6753/comments

https://api.github.com/repos/huggingface/datasets/issues/6753/events

https://github.com/huggingface/datasets/issues/6753

I_kwDODunzps6DYLzT

Type error when importing datasets on Kaggle

open

NONE

### Describe the bug When trying to run ``` import datasets print(datasets.__version__) ``` It generates the following error ``` TypeError: expected string or bytes-like object ``` It looks like It cannot find the valid versions of `fsspec` though fsspec version is fine when I checked Via command ``` import fsspec print(fsspec.__version__) ​ # output: 2024.3.1 ``` Detailed crash report ``` --------------------------------------------------------------------------- TypeError Traceback (most recent call last) Cell In[1], line 1 ----> 1 import datasets 2 print(datasets.__version__) File /opt/conda/lib/python3.10/site-packages/datasets/__init__.py:18 1 # ruff: noqa 2 # Copyright 2020 The HuggingFace Datasets Authors and the TensorFlow Datasets Authors. 3 # (...) 13 # See the License for the specific language governing permissions and 14 # limitations under the License. 16 __version__ = "2.18.0" ---> 18 from .arrow_dataset import Dataset 19 from .arrow_reader import ReadInstruction 20 from .builder import ArrowBasedBuilder, BeamBasedBuilder, BuilderConfig, DatasetBuilder, GeneratorBasedBuilder File /opt/conda/lib/python3.10/site-packages/datasets/arrow_dataset.py:66 63 from multiprocess import Pool 64 from tqdm.contrib.concurrent import thread_map ---> 66 from . import config 67 from .arrow_reader import ArrowReader 68 from .arrow_writer import ArrowWriter, OptimizedTypedSequence File /opt/conda/lib/python3.10/site-packages/datasets/config.py:41 39 # Imports 40 DILL_VERSION = version.parse(importlib.metadata.version("dill")) ---> 41 FSSPEC_VERSION = version.parse(importlib.metadata.version("fsspec")) 42 PANDAS_VERSION = version.parse(importlib.metadata.version("pandas")) 43 PYARROW_VERSION = version.parse(importlib.metadata.version("pyarrow")) File /opt/conda/lib/python3.10/site-packages/packaging/version.py:49, in parse(version) 43 """ 44 Parse the given version string and return either a :class:`Version` object 45 or a :class:`LegacyVersion` object depending on if the given version is 46 a valid PEP 440 version or a legacy version. 47 """ 48 try: ---> 49 return Version(version) 50 except InvalidVersion: 51 return LegacyVersion(version) File /opt/conda/lib/python3.10/site-packages/packaging/version.py:264, in Version.__init__(self, version) 261 def __init__(self, version: str) -> None: 262 263 # Validate the version and parse it into pieces --> 264 match = self._regex.search(version) 265 if not match: 266 raise InvalidVersion(f"Invalid version: '{version}'") TypeError: expected string or bytes-like object ``` ### Steps to reproduce the bug 1. run `!pip install -U datasets` on kaggle 2. check datasets is installed via ``` import datasets print(datasets.__version__) ``` ### Expected behavior Expected to print datasets version, like `2.18.0` ### Environment info Running on Kaggle, latest enviornment , here is the notebook https://www.kaggle.com/code/jtv199/mistrial-7b-part2

https://api.github.com/repos/huggingface/datasets/issues/6753/timeline

https://api.github.com/repos/huggingface/datasets/issues/6752

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6752/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6752/comments

https://api.github.com/repos/huggingface/datasets/issues/6752/events

https://github.com/huggingface/datasets/issues/6752

I_kwDODunzps6DXwo_

Precision being changed from float16 to float32 unexpectedly

open

NONE

### Describe the bug I'm loading a HuggingFace Dataset for images. I'm running a preprocessing (map operation) step that runs a few operations, one of them being conversion to float16. The Dataset features also say that the 'img' is of type float16. Whenever I take an image from that HuggingFace Dataset instance, the type turns out to be float32. ### Steps to reproduce the bug ```python import torchvision.transforms.v2 as transforms from datasets import load_dataset dataset = load_dataset('cifar10', split='test') dataset = dataset.with_format("torch") data_transform = transforms.Compose([transforms.Resize((32, 32)), transforms.ToDtype(torch.float16, scale=True), transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5]), ]) def _preprocess(examples): # Permutes from (BS x H x W x C) to (BS x C x H x W) images = torch.permute(examples['img'], (0, 3, 2, 1)) examples['img'] = data_transform(images) return examples dataset = dataset.map(_preprocess, batched=True, batch_size=8) ``` Now at this point the dataset.features are showing float16 which is great because that's what I want. ```python print(data_loader.features['img']) Sequence(feature=Sequence(feature=Sequence(feature=Value(dtype='float16', id=None), length=-1, id=None), length=-1, id=None), length=-1, id=None) ``` But when I try to sample an image from this dataloader; I'm getting a float32 image, when I'm expecting float16: ```python print(next(iter(data_loader))['img'].dtype) torch.float32 ``` ### Expected behavior I'm expecting the images loaded after the transformation to stay in float16. ### Environment info - `datasets` version: 2.18.0 - Platform: Linux-5.15.146.1-microsoft-standard-WSL2-x86_64-with-glibc2.31 - Python version: 3.10.9 - `huggingface_hub` version: 0.21.4 - PyArrow version: 14.0.2 - Pandas version: 2.0.3 - `fsspec` version: 2023.10.0

https://api.github.com/repos/huggingface/datasets/issues/6752/timeline

https://api.github.com/repos/huggingface/datasets/issues/6751

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6751/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6751/comments

https://api.github.com/repos/huggingface/datasets/issues/6751/events

https://github.com/huggingface/datasets/pull/6751

PR_kwDODunzps5qkKLH

Use 'with' operator for some download functions

closed

NONE

Some functions in `streaming_download_manager.py` are not closing the file they open which lead to `Unclosed file` warnings in our code. This fixes a few of them.

https://api.github.com/repos/huggingface/datasets/issues/6751/timeline

https://api.github.com/repos/huggingface/datasets/issues/6750

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6750/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6750/comments

https://api.github.com/repos/huggingface/datasets/issues/6750/events

https://github.com/huggingface/datasets/issues/6750

I_kwDODunzps6DWCAC

`load_dataset` requires a network connection for local download?

open

NONE

### Describe the bug Hi all - I see that in the past a network dependency has been mistakenly introduced into `load_dataset` even for local loads. Is it possible this has happened again? ### Steps to reproduce the bug ``` >>> import datasets >>> datasets.load_dataset("hh-rlhf") Repo card metadata block was not found. Setting CardData to empty. *hangs bc i'm firewalled* ```` stack trace from ctrl-c: ``` ^CTraceback (most recent call last): File "<stdin>", line 1, in <module> File "/home/jobuser/.local/lib/python3.10/site-packages/datasets/load.py", line 2582, in load_dataset builder_instance.download_and_prepare( output_path = get_from_cache( [0/122] File "/home/jobuser/.local/lib/python3.10/site-packages/datasets/utils/file_utils.py", line 532, in get_from_cache response = http_head( File "/home/jobuser/.local/lib/python3.10/site-packages/datasets/utils/file_utils.py", line 419, in http_head response = _request_with_retry( File "/home/jobuser/.local/lib/python3.10/site-packages/datasets/utils/file_utils.py", line 304, in _request_with_retry response = requests.request(method=method.upper(), url=url, timeout=timeout, **params) File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/requests/api.py", line 59, in request return session.request(method=method, url=url, **kwargs) File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/requests/sessions.py", line 587, in request resp = self.send(prep, **send_kwargs) File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/requests/sessions.py", line 701, in send r = adapter.send(request, **kwargs) File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/requests/adapters.py", line 487, in send resp = conn.urlopen( File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connectionpool.py", line 703, in urlopen httplib_response = self._make_request( File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connectionpool.py", line 386, in _make_request self._validate_conn(conn) File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connectionpool.py", line 1042, in _validate_conn conn.connect() File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connection.py", line 363, in connect self.sock = conn = self._new_conn() File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connection.py", line 174, in _new_conn conn = connection.create_connection( File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/util/connection.py", line 85, in create_connection sock.connect(sa) KeyboardInterrupt ``` ### Expected behavior loads the dataset ### Environment info ``` > pip show datasets Name: datasets Version: 2.18.0 ``` Python 3.10.2

https://api.github.com/repos/huggingface/datasets/issues/6750/timeline

https://api.github.com/repos/huggingface/datasets/issues/6749

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6749/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6749/comments

https://api.github.com/repos/huggingface/datasets/issues/6749/events

https://github.com/huggingface/datasets/pull/6749

PR_kwDODunzps5qeoSk

Fix fsspec tqdm callback

closed

MEMBER

Following changes at https://github.com/fsspec/filesystem_spec/pull/1497 for `fsspec>=2024.2.0`

https://api.github.com/repos/huggingface/datasets/issues/6749/timeline

https://api.github.com/repos/huggingface/datasets/issues/6748

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6748/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6748/comments

https://api.github.com/repos/huggingface/datasets/issues/6748/events

https://github.com/huggingface/datasets/issues/6748

I_kwDODunzps6DOH0k

Strange slicing behavior

open

NONE

### Describe the bug I have loaded a dataset, and then slice first 300 samples using `:` ops, however, the resulting dataset is not expected, as the output below: ```bash len(dataset)=1050324 len(dataset[:300])=2 len(dataset[0:300])=2 len(dataset.select(range(300)))=300 ``` ### Steps to reproduce the bug load a dataset then: ```bash dataset = load_from_disk(args.train_data_dir) print(f"{len(dataset)=}", flush=True) print(f"{len(dataset[:300])=}", flush=True) print(f"{len(dataset[0:300])=}", flush=True) print(f"{len(dataset.select(range(300)))=}", flush=True) ``` ### Expected behavior ```bash len(dataset)=1050324 len(dataset[:300])=300 len(dataset[0:300])=300 len(dataset.select(range(300)))=300 ``` ### Environment info - `datasets` version: 2.16.1 - Platform: Linux-5.15.0-60-generic-x86_64-with-glibc2.35 - Python version: 3.10.11 - `huggingface_hub` version: 0.20.2 - PyArrow version: 10.0.1 - Pandas version: 1.5.3 - `fsspec` version: 2023.10.0

https://api.github.com/repos/huggingface/datasets/issues/6748/timeline

https://api.github.com/repos/huggingface/datasets/issues/6747

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6747/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6747/comments

https://api.github.com/repos/huggingface/datasets/issues/6747/events

https://github.com/huggingface/datasets/pull/6747

PR_kwDODunzps5qa5L-

chore(deps): bump fsspec

closed

CONTRIBUTOR

There were a few fixes released recently, some DVC ecosystem packages require newer version of `fsspec`.

https://api.github.com/repos/huggingface/datasets/issues/6747/timeline

https://api.github.com/repos/huggingface/datasets/issues/6746

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6746/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6746/comments

https://api.github.com/repos/huggingface/datasets/issues/6746/events

https://github.com/huggingface/datasets/issues/6746

I_kwDODunzps6DEfwd

ExpectedMoreSplits error when loading C4 dataset

open

NONE

### Describe the bug I encounter bug when running the example command line ```python python main.py \ --model decapoda-research/llama-7b-hf \ --prune_method wanda \ --sparsity_ratio 0.5 \ --sparsity_type unstructured \ --save out/llama_7b/unstructured/wanda/ ``` The bug occurred at these lines of code (when loading c4 dataset) ```python 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') ``` The error message states: ``` raise ExpectedMoreSplits(str(set(expected_splits) - set(recorded_splits))) datasets.utils.info_utils.ExpectedMoreSplits: {'validation'} ``` ### Steps to reproduce the bug 1. I encounter bug when running the example command line ### Expected behavior The error message states: ``` raise ExpectedMoreSplits(str(set(expected_splits) - set(recorded_splits))) datasets.utils.info_utils.ExpectedMoreSplits: {'validation'} ``` ### Environment info I'm using cuda 12.4, so I use ```pip install pytorch``` instead of conda provided in install.md Also, I've tried another environment using the same commands in install.md, but the same bug occured

https://api.github.com/repos/huggingface/datasets/issues/6746/timeline

https://api.github.com/repos/huggingface/datasets/issues/6745

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6745/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6745/comments

https://api.github.com/repos/huggingface/datasets/issues/6745/events

https://github.com/huggingface/datasets/issues/6745

I_kwDODunzps6DCxWk

Scraping the whole of github including private repos is bad; kindly stop

closed

NONE

### Feature request https://github.com/bigcode-project/opt-out-v2 - opt out is not consent. kindly quit this ridiculous nonsense. ### Motivation [EDITED: insults not tolerated] ### Your contribution [EDITED: insults not tolerated]

https://api.github.com/repos/huggingface/datasets/issues/6745/timeline

completed

https://api.github.com/repos/huggingface/datasets/issues/6744

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6744/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6744/comments

https://api.github.com/repos/huggingface/datasets/issues/6744/events

https://github.com/huggingface/datasets/issues/6744

I_kwDODunzps6DAXKY

Option to disable file locking

open

NONE

### Feature request Commands such as `load_dataset` creates file locks with `filelock.FileLock`. It would be good if there was a way to disable this. ### Motivation File locking doesn't work on all file-systems (in my case NFS mounted Weka). If the `cache_dir` only had small files then it would be possible to point to local disk and the problem would be solved. However, as cache_dir is both where the small info files are written and the processed datasets are put this isn't a feasible solution. Considering https://github.com/huggingface/datasets/issues/6395 I still do think this is something that belongs in HuggingFace. The possibility to control packages separately is valuable. It might be that a user has their dataset on a file-system that doesn't support file-locking while they are using file locking on local disk to control some other type of access. ### Your contribution My suggested solution: ``` diff --git a/src/datasets/utils/_filelock.py b/src/datasets/utils/_filelock.py index 19620e6e..58f41a02 100644 --- a/src/datasets/utils/_filelock.py +++ b/src/datasets/utils/_filelock.py @@ -18,11 +18,15 @@ import os from filelock import FileLock as FileLock_ -from filelock import UnixFileLock +from filelock import SoftFileLock, UnixFileLock from filelock import __version__ as _filelock_version from packaging import version +if os.getenv('HF_USE_SOFTFILELOCK', 'false').lower() in ('true', '1'): + FileLock_ = SoftFileLock + + class FileLock(FileLock_): """ A `filelock.FileLock` initializer that handles long paths. ```

https://api.github.com/repos/huggingface/datasets/issues/6744/timeline

https://api.github.com/repos/huggingface/datasets/issues/6743

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6743/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6743/comments

https://api.github.com/repos/huggingface/datasets/issues/6743/events

https://github.com/huggingface/datasets/pull/6743

PR_kwDODunzps5qHeMZ

Allow null values in dict columns

closed

CONTRIBUTOR

Fix #6738

https://api.github.com/repos/huggingface/datasets/issues/6743/timeline

https://api.github.com/repos/huggingface/datasets/issues/6742

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6742/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6742/comments

https://api.github.com/repos/huggingface/datasets/issues/6742/events

https://github.com/huggingface/datasets/pull/6742

PR_kwDODunzps5qGSfG

Fix missing download_config in get_data_patterns

closed

MEMBER

Reported in https://github.com/huggingface/datasets-server/issues/2607

https://api.github.com/repos/huggingface/datasets/issues/6742/timeline

https://api.github.com/repos/huggingface/datasets/issues/6741

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6741/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6741/comments

https://api.github.com/repos/huggingface/datasets/issues/6741/events

https://github.com/huggingface/datasets/pull/6741

PR_kwDODunzps5qEiu3

Fix offline mode with single config

closed

MEMBER

Reported in https://github.com/huggingface/datasets/issues/4760 The cache was not able to reload a dataset with a single config form the cache if the config name is not specificed For example ```python from datasets import load_dataset, config config.HF_DATASETS_OFFLINE = True load_dataset("openai_humaneval") ``` This was due to a regression in https://github.com/huggingface/datasets/pull/6632

https://api.github.com/repos/huggingface/datasets/issues/6741/timeline

https://api.github.com/repos/huggingface/datasets/issues/6740

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6740/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6740/comments

https://api.github.com/repos/huggingface/datasets/issues/6740/events

https://github.com/huggingface/datasets/issues/6740

I_kwDODunzps6CuSZq

Support for loading geotiff files as a part of the ImageFolder

open

NONE

### Feature request Request for adding rasterio support to load geotiff as a part of ImageFolder, instead of using PIL ### Motivation As of now, there are many datasets in HuggingFace Hub which are predominantly focussed towards RemoteSensing or are from RemoteSensing. The current ImageFolder (if I have understood correctly) uses PIL. This is not really optimized because mostly these datasets have images with many channels and additional metadata. Using PIL makes one loose it unless we provide a custom script. Hence, maybe an API could be added to have this in common? ### Your contribution If the issue is accepted - i can contribute the code, because I would like to have it automated and generalised.

https://api.github.com/repos/huggingface/datasets/issues/6740/timeline

https://api.github.com/repos/huggingface/datasets/issues/6739

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6739/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6739/comments

https://api.github.com/repos/huggingface/datasets/issues/6739/events

https://github.com/huggingface/datasets/pull/6739

PR_kwDODunzps5p-Bwe

Transpose images with EXIF Orientation tag

closed

CONTRIBUTOR

Closes https://github.com/huggingface/datasets/issues/6252

https://api.github.com/repos/huggingface/datasets/issues/6739/timeline

https://api.github.com/repos/huggingface/datasets/issues/6738

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6738/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6738/comments

https://api.github.com/repos/huggingface/datasets/issues/6738/events

https://github.com/huggingface/datasets/issues/6738

I_kwDODunzps6CrSno

Dict feature is non-nullable while nested dict feature is

closed

CONTRIBUTOR

When i try to create a `Dataset` object with None values inside a dict column, like this: ```python from datasets import Dataset, Features, Value Dataset.from_dict( { "dict": [{"a": 0, "b": 0}, None], }, features=Features( {"dict": {"a": Value("int16"), "b": Value("int16")}} ) ) ``` i get `ValueError: Got None but expected a dictionary instead`. At the same time, having None in _nested_ dict feature works, for example, this doesn't throw any errors: ```python from datasets import Dataset, Features, Value, Sequence dataset = Dataset.from_dict( { "list_dict": [[{"a": 0, "b": 0}], None], "sequence_dict": [[{"a": 0, "b": 0}], None], }, features=Features({ "list_dict": [{"a": Value("int16"), "b": Value("int16")}], "sequence_dict": Sequence({"a": Value("int16"), "b": Value("int16")}), }) ) ``` Other types of features also seem to be nullable (but I haven't checked all of them). Version of `datasets` is the latest atm (2.18.0) Is this an expected behavior or a bug?

https://api.github.com/repos/huggingface/datasets/issues/6738/timeline

completed

https://api.github.com/repos/huggingface/datasets/issues/6737

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6737/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6737/comments

https://api.github.com/repos/huggingface/datasets/issues/6737/events

https://github.com/huggingface/datasets/issues/6737

I_kwDODunzps6Ci8aZ

Invalid pattern: '**' can only be an entire path component

open

NONE

### Describe the bug ValueError: Invalid pattern: '**' can only be an entire path component when loading any dataset ### Steps to reproduce the bug import datasets ds = datasets.load_dataset("TokenBender/code_instructions_122k_alpaca_style") ### Expected behavior loading the dataset successfully ### Environment info - `datasets` version: 2.18.0 - Platform: Windows-10-10.0.22631-SP0 - Python version: 3.11.7 - `huggingface_hub` version: 0.20.3 - PyArrow version: 15.0.0 - Pandas version: 2.2.1 - `fsspec` version: 2023.12.2

https://api.github.com/repos/huggingface/datasets/issues/6737/timeline

https://api.github.com/repos/huggingface/datasets/issues/6736

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6736/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6736/comments

https://api.github.com/repos/huggingface/datasets/issues/6736/events

https://github.com/huggingface/datasets/issues/6736

I_kwDODunzps6Ci4Qu

Mosaic Streaming (MDS) Support

open

NONE

### Feature request I'm a huge fan of the current HF Datasets `webdataset` integration (especially the built-in streaming support). However, I'd love to upload some robotics and multimodal datasets I've processed for use with [Mosaic Streaming](https://docs.mosaicml.com/projects/streaming/en/stable/), specifically their [MDS Format](https://docs.mosaicml.com/projects/streaming/en/stable/fundamentals/dataset_format.html#mds). Because the shard files have similar semantics to WebDataset, I'm hoping that adding such support won't be too much trouble? ### Motivation One of the downsides with WebDataset is a lack of out-of-the-box determinism (especially for large-scale training and reproducibility), easy job resumption, and the ability to quickly debug / visualize individual examples. Mosaic Streaming provides a [great interface for this out of the box](https://docs.mosaicml.com/projects/streaming/en/stable/#key-features), so I'd love to see it supported in HF Datasets. ### Your contribution Happy to help test things / provide example data. Can potentially submit a PR if maintainers could point me to the necessary WebDataset logic / steps for adding a new streaming format!

https://api.github.com/repos/huggingface/datasets/issues/6736/timeline

https://api.github.com/repos/huggingface/datasets/issues/6735

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6735/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6735/comments

https://api.github.com/repos/huggingface/datasets/issues/6735/events

https://github.com/huggingface/datasets/pull/6735

PR_kwDODunzps5px84g

Add `mode` parameter to `Image` feature

closed

CONTRIBUTOR

Fix https://github.com/huggingface/datasets/issues/6675

https://api.github.com/repos/huggingface/datasets/issues/6735/timeline

https://api.github.com/repos/huggingface/datasets/issues/6734

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6734/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6734/comments

https://api.github.com/repos/huggingface/datasets/issues/6734/events

https://github.com/huggingface/datasets/issues/6734

I_kwDODunzps6CZNbm

Tokenization slows towards end of dataset

open

NONE

### Describe the bug Mapped tokenization slows down substantially towards end of dataset. train set started off very slow, caught up to 20k then tapered off til the end. what's particularly strange is that the tokenization crashed a few times before due to errors with invalid tokens somewhere or corrupted downloads, and the speed ups/downs consistently happened the same times ```bash Running tokenizer on dataset (num_proc=48): 0%| | 847000/881416735 [12:18<252:45:45, 967.72 examples/s] Running tokenizer on dataset (num_proc=48): 0%| | 848000/881416735 [12:19<224:16:10, 1090.66 examples/s] Running tokenizer on dataset (num_proc=48): 10%|▉ | 84964000/881416735 [3:48:00<11:21:34, 19476.01 examples/s] Running tokenizer on dataset (num_proc=48): 10%|▉ | 84967000/881416735 [3:48:00<12:04:01, 18333.79 examples/s] Running tokenizer on dataset (num_proc=48): 61%|██████ | 538631977/881416735 [13:46:40<27:50:04, 3420.84 examples/s] Running tokenizer on dataset (num_proc=48): 61%|██████ | 538632977/881416735 [13:46:40<23:48:20, 3999.77 examples/s] Running tokenizer on dataset (num_proc=48): 100%|█████████▉| 881365886/881416735 [38:30:19<04:34, 185.10 examples/s] Running tokenizer on dataset (num_proc=48): 100%|█████████▉| 881366886/881416735 [38:30:25<04:36, 180.57 examples/s] ``` and validation set as well ```bash Running tokenizer on dataset (num_proc=48): 90%|████████▉ | 41544000/46390354 [28:44<02:37, 30798.76 examples/s] Running tokenizer on dataset (num_proc=48): 90%|████████▉ | 41550000/46390354 [28:44<02:08, 37698.08 examples/s] Running tokenizer on dataset (num_proc=48): 96%|█████████▋| 44747422/46390354 [2:15:48<12:22:44, 36.87 examples/s] Running tokenizer on dataset (num_proc=48): 96%|█████████▋| 44747422/46390354 [2:16:00<12:22:44, 36.87 examples/s] ``` ### Steps to reproduce the bug using the following kwargs ```python with accelerator.main_process_first(): lm_datasets = tokenized_datasets.map( group_texts, batched=True, num_proc=48 load_from_cache_file=True, desc=f"Grouping texts in chunks of {block_size}", ) ``` running through slurm script ```bash #SBATCH --partition=gpu-nvidia-a100 #SBATCH --nodes=1 #SBATCH --ntasks=1 #SBATCH --gpus-per-task=8 #SBATCH --cpus-per-task=96 ``` using this dataset https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T ### Expected behavior Constant speed throughout ### Environment info - `datasets` version: 2.15.0 - Platform: Linux-5.15.0-1049-aws-x86_64-with-glibc2.10 - Python version: 3.8.18 - `huggingface_hub` version: 0.19.4 - PyArrow version: 14.0.1 - Pandas version: 2.0.3 - `fsspec` version: 2023.10.0

https://api.github.com/repos/huggingface/datasets/issues/6734/timeline

https://api.github.com/repos/huggingface/datasets/issues/6733

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6733/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6733/comments

https://api.github.com/repos/huggingface/datasets/issues/6733/events

https://github.com/huggingface/datasets/issues/6733

I_kwDODunzps6CWBlM

EmptyDatasetError when loading dataset downloaded with HuggingFace cli

open

NONE

### Describe the bug I am using a cluster that does not have access to the internet when given a job. I tried downloading the dataset using the huggingface-cli command and then loading it with load_dataset but I get an error: ```raise EmptyDatasetError(f"The directory at {base_path} doesn't contain any data files") from None``` The dataset I'm using is "lmsys/chatbot_arena_conversations". The folder structure is - README.md - data - train-00000-of-00001-cced8514c7ed782a.parquet ### Steps to reproduce the bug 1. Download dataset using HuggingFace CLI: ```huggingface-cli download lmsys/chatbot_arena_conversations --local-dir ./lmsys/chatbot_arena_conversations``` 2. In Python ``` from datasets import load_dataset load_dataset("lmsys/chatbot_arena_conversations") ``` ### Expected behavior Should return a Dataset Dict in the form of ``` DatasetDict({ train: Dataset({ features: [...], num_rows: 33,000 }) }) ``` ### Environment info Python 3.11.5 Datasets 2.18.0 Transformers 4.38.2 Pytorch 2.2.0 Pyarrow 15.0.1 Rocky Linux release 8.9 (Green Obsidian)

https://api.github.com/repos/huggingface/datasets/issues/6733/timeline

https://api.github.com/repos/huggingface/datasets/issues/6731

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6731/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6731/comments

https://api.github.com/repos/huggingface/datasets/issues/6731/events

https://github.com/huggingface/datasets/issues/6731

I_kwDODunzps6CG5EB

Unexpected behavior when using load_dataset with streaming=True in a for loop

open

NONE

### Describe the bug ### My Code ``` from datasets import load_dataset res=[] for i in [0,1]: di=load_dataset( "json", data_files='path_to.json', split='train', streaming=True, ).map(lambda x: {"source": i}) res.append(di) for e in res[0]: print(e) ``` ### Unexpected Behavior Data in `res[0]` has `source=1`. However the expected value is 0. ### FYI I further switch `streaming` to `False`. And the output value is as expected (0). So there may exist bugs in setting `streaming=True` in a for loop. ### Environment Python 3.8.0 datasets==2.18.0 transformers==4.28.1 ### Steps to reproduce the bug 1. Create a Json file with any content. 2. Run the provided code. 3. Switch `streaming` to `False` and run again to see the expected behavior. ### Expected behavior The expected behavior is the data are mapped with its corresponding value in the for loop. ### Environment info Python 3.8.0 datasets==2.18.0 transformers==4.28.1 Ubuntu 20.04

https://api.github.com/repos/huggingface/datasets/issues/6731/timeline

https://api.github.com/repos/huggingface/datasets/issues/6730

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6730/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6730/comments

https://api.github.com/repos/huggingface/datasets/issues/6730/events

https://github.com/huggingface/datasets/pull/6730

PR_kwDODunzps5pZDsB

Deprecate Pandas builder

closed

CONTRIBUTOR

The Pandas packaged builder is undocumented and relies on `pickle` to read the data, making it **unsafe**. Moreover, I haven't seen a single instance of this builder being used (not even using the GH/Hub search), so we should deprecate it.

https://api.github.com/repos/huggingface/datasets/issues/6730/timeline

https://api.github.com/repos/huggingface/datasets/issues/6729

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6729/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6729/comments

https://api.github.com/repos/huggingface/datasets/issues/6729/events

https://github.com/huggingface/datasets/issues/6729

I_kwDODunzps6B88dn

Support zipfiles that span multiple disks?

open

CONTRIBUTOR

See https://huggingface.co/datasets/PhilEO-community/PhilEO-downstream The dataset viewer gives the following error: ``` Error code: ConfigNamesError Exception: BadZipFile Message: zipfiles that span multiple disks are not supported Traceback: Traceback (most recent call last): File "/src/services/worker/src/worker/job_runners/dataset/config_names.py", line 67, in compute_config_names_response get_dataset_config_names( File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/inspect.py", line 347, in get_dataset_config_names dataset_module = dataset_module_factory( File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 1871, in dataset_module_factory raise e1 from None File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 1846, in dataset_module_factory return HubDatasetModuleFactoryWithoutScript( File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 1240, in get_module module_name, default_builder_kwargs = infer_module_for_data_files( File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 584, in infer_module_for_data_files split_modules = { File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 585, in <dictcomp> split: infer_module_for_data_files_list(data_files_list, download_config=download_config) File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 526, in infer_module_for_data_files_list return infer_module_for_data_files_list_in_archives(data_files_list, download_config=download_config) File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 554, in infer_module_for_data_files_list_in_archives for f in xglob(extracted, recursive=True, download_config=download_config)[ File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/download/streaming_download_manager.py", line 576, in xglob fs, *_ = fsspec.get_fs_token_paths(urlpath, storage_options=storage_options) File "/src/services/worker/.venv/lib/python3.9/site-packages/fsspec/core.py", line 622, in get_fs_token_paths fs = filesystem(protocol, **inkwargs) File "/src/services/worker/.venv/lib/python3.9/site-packages/fsspec/registry.py", line 290, in filesystem return cls(**storage_options) File "/src/services/worker/.venv/lib/python3.9/site-packages/fsspec/spec.py", line 79, in __call__ obj = super().__call__(*args, **kwargs) File "/src/services/worker/.venv/lib/python3.9/site-packages/fsspec/implementations/zip.py", line 57, in __init__ self.zip = zipfile.ZipFile( File "/usr/local/lib/python3.9/zipfile.py", line 1266, in __init__ self._RealGetContents() File "/usr/local/lib/python3.9/zipfile.py", line 1329, in _RealGetContents endrec = _EndRecData(fp) File "/usr/local/lib/python3.9/zipfile.py", line 286, in _EndRecData return _EndRecData64(fpin, -sizeEndCentDir, endrec) File "/usr/local/lib/python3.9/zipfile.py", line 232, in _EndRecData64 raise BadZipFile("zipfiles that span multiple disks are not supported") zipfile.BadZipFile: zipfiles that span multiple disks are not supported ``` The files (https://huggingface.co/datasets/PhilEO-community/PhilEO-downstream/tree/main/data) are: <img width="629" alt="Capture d’écran 2024-03-11 à 22 07 30" src="https://github.com/huggingface/datasets/assets/1676121/0bb15a51-d54f-4d73-8572-e427ea644b36">

https://api.github.com/repos/huggingface/datasets/issues/6729/timeline

https://api.github.com/repos/huggingface/datasets/issues/6728

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6728/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6728/comments

https://api.github.com/repos/huggingface/datasets/issues/6728/events

https://github.com/huggingface/datasets/issues/6728

I_kwDODunzps6B2uek

Issue Downloading Certain Datasets After Setting Custom `HF_ENDPOINT`

closed

NONE

### Describe the bug This bug is triggered under the following conditions: - datasets repo ids without organization names trigger errors, such as `bookcorpus`, `gsm8k`, `wikipedia`, rather than in the form of `A/B`. - If `HF_ENDPOINT` is set and the hostname is not in the form of `(hub-ci.)?huggingface.co`. - This issue occurs with `datasets>2.15.0` or `huggingface-hub>0.19.4`. For example, using the latest versions: `datasets==2.18.0` and `huggingface-hub==0.21.4`, ### Steps to reproduce the bug the issue can be reproduced with the following code: 1. install specific datasets and huggingface_hub. ```bash pip install datasets==2.18.0 pip install huggingface_hub==0.21.4 ``` 2. execute python code. ```Python import os os.environ['HF_ENDPOINT'] = 'https://hf-mirror.com' from datasets import load_dataset bookcorpus = load_dataset('bookcorpus', split='train') ``` console output: ``` Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/home/padeoe/.local/lib/python3.10/site-packages/datasets/load.py", line 2556, in load_dataset builder_instance = load_dataset_builder( File "/home/padeoe/.local/lib/python3.10/site-packages/datasets/load.py", line 2228, in load_dataset_builder dataset_module = dataset_module_factory( File "/home/padeoe/.local/lib/python3.10/site-packages/datasets/load.py", line 1879, in dataset_module_factory raise e1 from None File "/home/padeoe/.local/lib/python3.10/site-packages/datasets/load.py", line 1830, in dataset_module_factory with fs.open(f"datasets/{path}/{filename}", "r", encoding="utf-8") as f: File "/home/padeoe/.local/lib/python3.10/site-packages/fsspec/spec.py", line 1295, in open self.open( File "/home/padeoe/.local/lib/python3.10/site-packages/fsspec/spec.py", line 1307, in open f = self._open( File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_file_system.py", line 228, in _open return HfFileSystemFile(self, path, mode=mode, revision=revision, block_size=block_size, **kwargs) File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_file_system.py", line 615, in __init__ self.resolved_path = fs.resolve_path(path, revision=revision) File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_file_system.py", line 180, in resolve_path repo_and_revision_exist, err = self._repo_and_revision_exist(repo_type, repo_id, revision) File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_file_system.py", line 117, in _repo_and_revision_exist self._api.repo_info(repo_id, revision=revision, repo_type=repo_type) File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 118, in _inner_fn return fn(*args, **kwargs) File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_api.py", line 2413, in repo_info return method( File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 118, in _inner_fn return fn(*args, **kwargs) File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_api.py", line 2286, in dataset_info hf_raise_for_status(r) File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/utils/_errors.py", line 362, in hf_raise_for_status raise HfHubHTTPError(str(e), response=response) from e huggingface_hub.utils._errors.HfHubHTTPError: 401 Client Error: Unauthorized for url: https://hf-mirror.com/api/datasets/bookcorpus/bookcorpus.py (Request ID: Root=1-65ee8659-5ab10eec5960c63e71f2bb58;b00bdbea-fd6e-4a74-8fe0-bc4682ae090e) ``` ### Expected behavior The dataset was downloaded correctly without any errors. ### Environment info datasets==2.18.0 huggingface-hub==0.21.4

https://api.github.com/repos/huggingface/datasets/issues/6728/timeline

completed

https://api.github.com/repos/huggingface/datasets/issues/6727

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6727/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6727/comments

https://api.github.com/repos/huggingface/datasets/issues/6727/events

https://github.com/huggingface/datasets/pull/6727

PR_kwDODunzps5pLJyE

Using a registry instead of calling globals for fetching feature types

closed

CONTRIBUTOR

Hello, When working with bio-data, each feature often has metadata associated with it (e.g. species, lineage, snp position, etc). To store this, I like to use the feature classes with the added `metadata` attribute. However, when saving or loading with custom features, you get an error since that class doesn't exist in the global namespace in `datasets.features.features`. Take for example, ```python from dataclasses import dataclass, field from datasets import Dataset from datasets.features.features import Value, Features @dataclass class FeatureA(Value): metadata: dict = field(default=dict) _type: str = field(default="FeatureA", init=False, repr=False) @dataclass class FeatureB(Value): metadata: dict = field(default_factory=dict) _type: str = field(default="FeatureB", init=False, repr=False) test_data = { "a": [1, 2, 3], "b": [4, 5, 6], } test_data = Dataset.from_dict( test_data, features=Features({ "a": FeatureA("int32", metadata={"species": "lactobacillus acetotolerans"}), "b": FeatureB("int32", metadata={"species": "lactobacillus iners"}), }) ) # returns an error since FeatureA and FeatureB are not in the global namespace test_data.save_to_disk('./test_data') ``` Saving the dataset (0/1 shards): 0%| | 0/3 [00:00<?, ? examples/s] --------------------------------------------------------------------------- KeyError Traceback (most recent call last) Cell In[2], line 28 19 test_data = Dataset.from_dict( 20 test_data, 21 features=Features({ (...) 24 }) 25 ) 27 # returns an error since FeatureA and FeatureB are not in the global namespace ---> 28 test_data.save_to_disk('./test_data') ... File ~\Documents\datasets\src\datasets\features\features.py:1361, in generate_from_dict(obj) 1359 return {key: generate_from_dict(value) for key, value in obj.items()} 1360 obj = dict(obj) -> 1361 class_type = globals()[obj.pop("_type")] 1363 if class_type == Sequence: 1364 return Sequence(feature=generate_from_dict(obj["feature"]), length=obj.get("length", -1)) KeyError: 'FeatureA' We can avoid this by having a registry (like formatters) and doing ```python from datasets.features.features import register_feature register_feature(FeatureA, "FeatureA") register_feature(FeatureB, "FeatureB") test_data.save_to_disk('./test_data') ``` Saving the dataset (1/1 shards): 100%|------| 3/3 [00:00<00:00, 211.13 examples/s] and loading from disk returns with all metadata information ```python from datasets import load_from_disk test_data = load_from_disk('./test_data') test_data.features ``` {'a': FeatureA(dtype='int32', id=None, metadata={'species': 'lactobacillus acetotolerans'}), 'b': FeatureB(dtype='int32', id=None, metadata={'species': 'lactobacillus iners'})}

https://api.github.com/repos/huggingface/datasets/issues/6727/timeline

https://api.github.com/repos/huggingface/datasets/issues/6726

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6726/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6726/comments

https://api.github.com/repos/huggingface/datasets/issues/6726/events

https://github.com/huggingface/datasets/issues/6726

I_kwDODunzps6Bw94Q

Profiling for HF Filesystem shows there are easy performance gains to be made

open

NONE

### Describe the bug # Let's make it faster First, an evidence...  Figure 1: CProfile for loading 3 files from cerebras/SlimPajama-627B train split, and 3 files from test split using streaming=True. X axis is 1106 seconds long. See? It's pretty slow. What is resolve pattern doing? ``` resolve_pattern called with **/train/** and hf://datasets/cerebras/SlimPajama-627B@2d0accdd58c5d5511943ca1f5ff0e3eb5e293543 resolve_pattern took 20.815081119537354 seconds ``` Makes sense. How to improve it? ## Bigger project, biggest payoff Databricks (and consequently, spark) store a compressed manifest file of the files contained in the remote filesystem. Then, you download one tiny file, decompress it, and all the operations are local instead of this shenanigans. It seems pretty straightforward to make dataset uploads compute a manifest and upload it alongside their data. This would make resolution time so fast that nobody would ever think about it again. It also means you either need to have the uploader compute it _every time_, or have a hook that computes it. ## Smaller project, immediate payoff: Be diligent in avoiding deepcopy Revise the _ls_tree method to avoid deepcopy: ``` def _ls_tree( self, path: str, recursive: bool = False, refresh: bool = False, revision: Optional[str] = None, expand_info: bool = True, ): ..... omitted ..... for path_info in tree: if isinstance(path_info, RepoFile): cache_path_info = { "name": root_path + "/" + path_info.path, "size": path_info.size, "type": "file", "blob_id": path_info.blob_id, "lfs": path_info.lfs, "last_commit": path_info.last_commit, "security": path_info.security, } else: cache_path_info = { "name": root_path + "/" + path_info.path, "size": 0, "type": "directory", "tree_id": path_info.tree_id, "last_commit": path_info.last_commit, } parent_path = self._parent(cache_path_info["name"]) self.dircache.setdefault(parent_path, []).append(cache_path_info) out.append(cache_path_info) return copy.deepcopy(out) # copy to not let users modify the dircache ``` Observe this deepcopy at the end. It is making a copy of a very simple data structure. We do not need to copy. We can simply generate the data structure twice instead. It will be much faster. ``` def _ls_tree( self, path: str, recursive: bool = False, refresh: bool = False, revision: Optional[str] = None, expand_info: bool = True, ): ..... omitted ..... def make_cache_path_info(path_info): if isinstance(path_info, RepoFile): return { "name": root_path + "/" + path_info.path, "size": path_info.size, "type": "file", "blob_id": path_info.blob_id, "lfs": path_info.lfs, "last_commit": path_info.last_commit, "security": path_info.security, } else: return { "name": root_path + "/" + path_info.path, "size": 0, "type": "directory", "tree_id": path_info.tree_id, "last_commit": path_info.last_commit, } for path_info in tree: cache_path_info = make_cache_path_info(path_info) out_cache_path_info = make_cache_path_info(path_info) # copy to not let users modify the dircache parent_path = self._parent(cache_path_info["name"]) self.dircache.setdefault(parent_path, []).append(cache_path_info) out.append(out_cache_path_info) return out ``` Note there is no longer a deepcopy in this method. We have replaced it with generating the output twice. This is substantially faster. For me, the entire resolution went from 1100s to 360s. ## Medium project, medium payoff After the above change, we have this profile:  Figure 2: x-axis is 355 seconds. Note that globbing and _ls_tree deep copy is gone. No surprise there. It's much faster now, but we still spend ~187seconds in get_fs_token_paths. Well get_fs_token_paths is part of fsspec. We don't need to fix that because we can trust their developers to write high performance code. Probably the caller has misconfigured something. Let's take a look at the storage_options being provided to the filesystem that is constructed during this call. Ah yes, streaming_download_manager::_prepare_single_hop_path_and_storage_options. We know streaming download manager is not compatible with async right now, but we really need this specific part of the code to be async. We're spending so much time checking isDir on the remote filesystem, it's a huge waste. We can make the call easily 20-30x faster by using async, removing this performance bottleneck almost entirely (and reducing the total time of this part of the code to <30s. There is no reason to block async isDir calls for streaming. I'm not going to mess w/ this one myself; I didn't write the streaming impl, and I don't know how it works, but I know the isDir check can be async. ### Steps to reproduce the bug ``` with cProfile.Profile() as pr: pr.enable() # Begin Data if not os.path.exists(data_cache_dir): os.makedirs(data_cache_dir, exist_ok=True) training_dataset = load_dataset(training_dataset_name, split=training_split, cache_dir=data_cache_dir, streaming=True).take(training_slice) eval_dataset = load_dataset(eval_dataset_name, split=eval_split, cache_dir=data_cache_dir, streaming=True).take(eval_slice) # End Data pr.disable() pr.create_stats() if not os.path.exists(profiling_path): os.makedirs(profiling_path, exist_ok=True) pr.dump_stats(os.path.join(profiling_path, "cprofile.prof")) ``` run this code for "cerebras/SlimPajama-627B" and whatever other params ### Expected behavior Something better. ### Environment info - `datasets` version: 2.18.0 - Platform: Linux-5.15.146.1-microsoft-standard-WSL2-x86_64-with-glibc2.35 - Python version: 3.10.13 - `huggingface_hub` version: 0.21.3 - PyArrow version: 15.0.0 - Pandas version: 2.2.1 - `fsspec` version: 2024.2.0

https://api.github.com/repos/huggingface/datasets/issues/6726/timeline

https://api.github.com/repos/huggingface/datasets/issues/6725

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6725/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6725/comments

https://api.github.com/repos/huggingface/datasets/issues/6725/events

https://github.com/huggingface/datasets/issues/6725

I_kwDODunzps6Bq-pq

Request for a comparison of huggingface datasets compared with other data format especially webdataset

open

NONE

### Feature request Request for a comparison of huggingface datasets compared with other data format especially webdataset ### Motivation I see huggingface datasets uses Apache Arrow as its backend, it seems to be great, but I'm curious about how it is good compared with other dataset format, like webdataset, what's the pros/cons of them. ### Your contribution More information

https://api.github.com/repos/huggingface/datasets/issues/6725/timeline

https://api.github.com/repos/huggingface/datasets/issues/6724

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6724/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6724/comments

https://api.github.com/repos/huggingface/datasets/issues/6724/events

https://github.com/huggingface/datasets/issues/6724

I_kwDODunzps6Bmq8T

Dataset with loading script does not work in renamed repos

open

CONTRIBUTOR

### Describe the bug My data repository was first called `BramVanroy/hplt-mono-v1-2` but I then renamed to use underscores instead of dashes. However, it seems that `datasets` retrieves the old repo name when it checks whether the repo contains data loading scripts in this line. https://github.com/huggingface/datasets/blob/6fb6c834f008996c994b0a86c3808d0a33d44525/src/datasets/load.py#L1845 When I print `filename` it returns `hplt-mono-v1-2.py` but the files in the repo are of course `['.gitattributes', 'README.md', 'hplt_mono_v1_2.py']`. So the `filename` is the original reponame instead of the renamed one. I am not sure if this is a caching issue or not or how I can resolve it. ### Steps to reproduce the bug ``` from datasets import load_dataset ds = load_dataset( "BramVanroy/hplt-mono-v1-2", "ky", trust_remote_code=True ) ``` ### Expected behavior That the most recent repo name is used when `filename` is generated. ### Environment info - `datasets` version: 2.16.1 - Platform: Linux-5.14.0-284.25.1.el9_2.x86_64-x86_64-with-glibc2.34 - Python version: 3.10.13 - `huggingface_hub` version: 0.20.2 - PyArrow version: 14.0.1 - Pandas version: 2.1.3 - `fsspec` version: 2023.10.0

https://api.github.com/repos/huggingface/datasets/issues/6724/timeline

https://api.github.com/repos/huggingface/datasets/issues/6723

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6723/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6723/comments

https://api.github.com/repos/huggingface/datasets/issues/6723/events

https://github.com/huggingface/datasets/pull/6723

PR_kwDODunzps5o_fPU

get_dataset_default_config_name docstring

closed

MEMBER

fix https://github.com/huggingface/datasets/pull/6722

https://api.github.com/repos/huggingface/datasets/issues/6723/timeline

https://api.github.com/repos/huggingface/datasets/issues/6722

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6722/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6722/comments

https://api.github.com/repos/huggingface/datasets/issues/6722/events

https://github.com/huggingface/datasets/pull/6722

PR_kwDODunzps5o_ch0

Add details in docstring

closed

CONTRIBUTOR

see https://github.com/huggingface/datasets-server/pull/2554#discussion_r1516516867

https://api.github.com/repos/huggingface/datasets/issues/6722/timeline

https://api.github.com/repos/huggingface/datasets/issues/6721

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6721/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6721/comments

https://api.github.com/repos/huggingface/datasets/issues/6721/events

https://github.com/huggingface/datasets/issues/6721

I_kwDODunzps6Bk5DC

Hi,do you know how to load the dataset from local file now?

open

NONE

Hi, if I want to load the dataset from local file, then how to specify the configuration name? _Originally posted by @WHU-gentle in https://github.com/huggingface/datasets/issues/2976#issuecomment-1333455222_

https://api.github.com/repos/huggingface/datasets/issues/6721/timeline

https://api.github.com/repos/huggingface/datasets/issues/6720

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6720/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6720/comments

https://api.github.com/repos/huggingface/datasets/issues/6720/events

https://github.com/huggingface/datasets/issues/6720

I_kwDODunzps6Bjo6D

TypeError: 'str' object is not callable

closed

CONTRIBUTOR

### Describe the bug I am trying to get the HPLT datasets on the hub. Downloading/re-uploading would be too time- and resource consuming so I wrote [a dataset loader script](https://huggingface.co/datasets/BramVanroy/hplt_mono_v1_2/blob/main/hplt_mono_v1_2.py). I think I am very close but for some reason I always get the error below. It happens during the clean-up phase where the directory cannot be removed because it is not empty. My only guess would be that this may have to do with zstandard ``` Traceback (most recent call last): File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1744, in _prepare_split_single writer.write(example, key) File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 492, in write self.write_examples_on_file() File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 434, in write_examples_on_file if self.schema File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 409, in schema else (pa.schema(self._features.type) if self._features is not None else None) File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1643, in type return get_nested_type(self) File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1209, in get_nested_type {key: get_nested_type(schema[key]) for key in schema} File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1209, in <dictcomp> {key: get_nested_type(schema[key]) for key in schema} File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1221, in get_nested_type value_type = get_nested_type(schema.feature) File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1228, in get_nested_type return schema() TypeError: 'str' object is not callable During handling of the above exception, another exception occurred: Traceback (most recent call last): File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1753, in _prepare_split_single num_examples, num_bytes = writer.finalize() File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 588, in finalize self.write_examples_on_file() File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 434, in write_examples_on_file if self.schema File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 409, in schema else (pa.schema(self._features.type) if self._features is not None else None) File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1643, in type return get_nested_type(self) File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1209, in get_nested_type {key: get_nested_type(schema[key]) for key in schema} File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1209, in <dictcomp> {key: get_nested_type(schema[key]) for key in schema} File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1221, in get_nested_type value_type = get_nested_type(schema.feature) File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1228, in get_nested_type return schema() TypeError: 'str' object is not callable The above exception was the direct cause of the following exception: Traceback (most recent call last): File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 959, in incomplete_dir yield tmp_dir File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1005, in download_and_prepare self._download_and_prepare( File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1767, in _download_and_prepare super()._download_and_prepare( File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1100, in _download_and_prepare self._prepare_split(split_generator, **prepare_split_kwargs) File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1605, in _prepare_split for job_id, done, content in self._prepare_split_single( File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1762, in _prepare_split_single raise DatasetGenerationError("An error occurred while generating the dataset") from e datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset During handling of the above exception, another exception occurred: Traceback (most recent call last): File "/home/pricie/vanroy/.config/JetBrains/PyCharm2023.3/scratches/scratch_5.py", line 4, in <module> ds = load_dataset( File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/load.py", line 2549, in load_dataset builder_instance.download_and_prepare( File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 985, in download_and_prepare with incomplete_dir(self._output_dir) as tmp_output_dir: File "/home/pricie/vanroy/.pyenv/versions/3.10.13/lib/python3.10/contextlib.py", line 153, in __exit__ self.gen.throw(typ, value, traceback) File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 966, in incomplete_dir shutil.rmtree(tmp_dir) File "/home/pricie/vanroy/.pyenv/versions/3.10.13/lib/python3.10/shutil.py", line 731, in rmtree onerror(os.rmdir, path, sys.exc_info()) File "/home/pricie/vanroy/.pyenv/versions/3.10.13/lib/python3.10/shutil.py", line 729, in rmtree os.rmdir(path) OSError: [Errno 39] Directory not empty: '/home/pricie/vanroy/.cache/huggingface/datasets/BramVanroy___hplt_mono_v1_2/ky/1.2.0/7ab138629fe7e9e29fe93ce63d809d5ef9d963273b829f61ab538e012dc9cc47.incomplete' ``` Interestingly, though, this directory _does_ appear to be empty: ```shell > cd /home/pricie/vanroy/.cache/huggingface/datasets/BramVanroy___hplt_mono_v1_2/ky/1.2.0/7ab138629fe7e9e29fe93ce63d809d5ef9d963273b829f61ab538e012dc9cc47.incomplete > ls -lah total 0 drwxr-xr-x. 1 vanroy vanroy 0 Mar 7 12:01 . drwxr-xr-x. 1 vanroy vanroy 304 Mar 7 11:52 .. > cd .. > ls 7ab138629fe7e9e29fe93ce63d809d5ef9d963273b829f61ab538e012dc9cc47_builder.lock 7ab138629fe7e9e29fe93ce63d809d5ef9d963273b829f61ab538e012dc9cc47.incomplete ``` ### Steps to reproduce the bug ```python from datasets import load_dataset ds = load_dataset( "BramVanroy/hplt_mono_v1_2", "ky", trust_remote_code=True ) ``` ### Expected behavior No error. ### Environment info - `datasets` version: 2.16.1 - Platform: Linux-5.14.0-284.25.1.el9_2.x86_64-x86_64-with-glibc2.34 - Python version: 3.10.13 - `huggingface_hub` version: 0.20.2 - PyArrow version: 14.0.1 - Pandas version: 2.1.3 - `fsspec` version: 2023.10.0

https://api.github.com/repos/huggingface/datasets/issues/6720/timeline

completed

https://api.github.com/repos/huggingface/datasets/issues/6719

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6719/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6719/comments

https://api.github.com/repos/huggingface/datasets/issues/6719/events

https://github.com/huggingface/datasets/issues/6719

I_kwDODunzps6BUUA_

Is there any way to solve hanging of IterableDataset using split by node + filtering during inference

open

NONE

### Describe the bug I am using an iterable dataset in a multi-node setup, trying to do training/inference while filtering the data on the fly. I usually do not use `split_dataset_by_node` but it is very slow using the IterableDatasetShard in `accelerate` and `transformers`. When I filter after applying `split_dataset_by_node`, it results in shards that are not equal sizes due to unequal samples filtered from each one. The distributed process hangs when trying to accomplish this. Is there any way to resolve this or is it impossible to implement? ### Steps to reproduce the bug Here is a toy example of what I am trying to do that reproduces the behavior ``` # torchrun --nproc-per-node 2 file.py import os import pandas as pd import torch from accelerate import Accelerator from datasets import Features, Value, load_dataset from datasets.distributed import split_dataset_by_node from torch.utils.data import DataLoader accelerator = Accelerator(device_placement=True, dispatch_batches=False) if accelerator.is_main_process: if not os.path.exists("scratch_data"): os.mkdir("scratch_data") n_shards = 4 for i in range(n_shards): df = pd.DataFrame({"id": list(range(10 * i, 10 * (i + 1)))}) df.to_parquet(f"scratch_data/shard_{i}.parquet") world_size = accelerator.num_processes local_rank = accelerator.process_index def collate_fn(examples): input_ids = [] for example in examples: input_ids.append(example["id"]) return torch.LongTensor(input_ids) dataset = load_dataset( "parquet", data_dir="scratch_data", split="train", streaming=True ) dataset = ( split_dataset_by_node(dataset, rank=local_rank, world_size=world_size) .filter(lambda x: x["id"] < 35) .shuffle(seed=42, buffer_size=100) ) batch_size = 2 train_dataloader = DataLoader( dataset, batch_size=batch_size, collate_fn=collate_fn, num_workers=2 ) for x in train_dataloader: x = x.to(accelerator.device) print({"rank": local_rank, "id": x}) y = accelerator.gather_for_metrics(x) if accelerator.is_main_process: print("gathered", y) ``` ### Expected behavior Is there any way to continue training/inference on the GPUs that have remaining data left without waiting for the others? Is it impossible to filter when ### Environment info - `datasets` version: 2.18.0 - Platform: Linux-5.10.209-198.812.amzn2.x86_64-x86_64-with-glibc2.31 - Python version: 3.10.13 - `huggingface_hub` version: 0.21.3 - PyArrow version: 15.0.0 - Pandas version: 2.2.1 - `fsspec` version: 2023.6.0

https://api.github.com/repos/huggingface/datasets/issues/6719/timeline

https://api.github.com/repos/huggingface/datasets/issues/6718

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6718/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6718/comments

https://api.github.com/repos/huggingface/datasets/issues/6718/events

https://github.com/huggingface/datasets/pull/6718

PR_kwDODunzps5ouwwE

Fix concurrent script loading with force_redownload

closed

MEMBER

I added `lock_importable_file` in `get_dataset_builder_class` and `extend_dataset_builder_for_streaming` to fix the issue, and I also added a test cc @clefourrier

https://api.github.com/repos/huggingface/datasets/issues/6718/timeline

https://api.github.com/repos/huggingface/datasets/issues/6717

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6717/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6717/comments

https://api.github.com/repos/huggingface/datasets/issues/6717/events

https://github.com/huggingface/datasets/issues/6717

I_kwDODunzps6BRCOg

`remove_columns` method used with a streaming enable dataset mode produces a LibsndfileError on multichannel audio

open

NONE

### Describe the bug When loading a HF dataset in streaming mode and removing some columns, it is impossible to load a sample if the audio contains more than one channel. I have the impression that the time axis and channels are swapped or concatenated. ### Steps to reproduce the bug Minimal error code: ```python from datasets import load_dataset dataset_name = "zinc75/Vibravox_dummy" config_name = "BWE_Larynx_microphone" # if we use "ASR_Larynx_microphone" subset which is a monochannel audio, no error is thrown. dataset = load_dataset( path=dataset_name, name=config_name, split="train", streaming=True ) dataset = dataset.remove_columns(["sensor_id"]) # dataset = dataset.map(lambda x:x, remove_columns=["sensor_id"]) # The commented version does not produce an error, but loses the dataset features. sample = next(iter(dataset)) ``` Error: ``` Traceback (most recent call last): File "/home/julien/Bureau/github/vibravox/tmp.py", line 15, in <module> sample = next(iter(dataset)) ^^^^^^^^^^^^^^^^^^^ File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/iterable_dataset.py", line 1392, in __iter__ example = _apply_feature_types_on_example( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/iterable_dataset.py", line 1080, in _apply_feature_types_on_example encoded_example = features.encode_example(example) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/features.py", line 1889, in encode_example return encode_nested_example(self, example) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/features.py", line 1244, in encode_nested_example {k: encode_nested_example(schema[k], obj.get(k), level=level + 1) for k in schema} File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/features.py", line 1244, in <dictcomp> {k: encode_nested_example(schema[k], obj.get(k), level=level + 1) for k in schema} ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/features.py", line 1300, in encode_nested_example return schema.encode_example(obj) if obj is not None else None ^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/audio.py", line 98, in encode_example sf.write(buffer, value["array"], value["sampling_rate"], format="wav") File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/soundfile.py", line 343, in write with SoundFile(file, 'w', samplerate, channels, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/soundfile.py", line 658, in __init__ self._file = self._open(file, mode_int, closefd) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/soundfile.py", line 1216, in _open raise LibsndfileError(err, prefix="Error opening {0!r}: ".format(self.name)) soundfile.LibsndfileError: Error opening <_io.BytesIO object at 0x7fd795d24680>: Format not recognised. Process finished with exit code 1 ``` ### Expected behavior I would expect this code to run without error. ### Environment info - `datasets` version: 2.18.0 - Platform: Linux-6.5.0-21-generic-x86_64-with-glibc2.35 - Python version: 3.11.0 - `huggingface_hub` version: 0.21.3 - PyArrow version: 15.0.0 - Pandas version: 2.2.1 - `fsspec` version: 2023.10.0

https://api.github.com/repos/huggingface/datasets/issues/6717/timeline

https://api.github.com/repos/huggingface/datasets/issues/6716

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6716/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6716/comments

https://api.github.com/repos/huggingface/datasets/issues/6716/events

https://github.com/huggingface/datasets/issues/6716

I_kwDODunzps6BQ9X-

Non-deterministic `Dataset.builder_name` value

closed

NONE

### Describe the bug I'm not sure if this is a bug, but `print(ds.builder_name)` in the following code sometimes prints out `rotten_tomatoes` instead of `parquet`: ```python import datasets for _ in range(100): ds = datasets.load_dataset("rotten_tomatoes", split="train") print(ds.builder_name) # prints out "rotten_tomatoes" sometimes instead of "parquet" ``` Output: ``` ... parquet parquet parquet rotten_tomatoes parquet parquet parquet ... ``` Here's a reproduction using GitHub Actions: https://github.com/mlflow/mlflow/actions/runs/8153247984/job/22284263613?pr=11329#step:12:241 One of our tests is flaky because `builder_name` is not deterministic. ### Steps to reproduce the bug 1. Run the code above. ### Expected behavior Always prints out `parquet`? ### Environment info ``` Copy-and-paste the text below in your GitHub issue. - `datasets` version: 2.18.0 - Platform: Linux-6.5.0-1015-azure-x86_64-with-glibc2.34 - Python version: 3.8.18 - `huggingface_hub` version: 0.21.3 - PyArrow version: 15.0.0 - Pandas version: 2.0.3 - `fsspec` version: 2024.2.0 ```

https://api.github.com/repos/huggingface/datasets/issues/6716/timeline

completed

https://api.github.com/repos/huggingface/datasets/issues/6715

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6715/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6715/comments

https://api.github.com/repos/huggingface/datasets/issues/6715/events

https://github.com/huggingface/datasets/pull/6715

PR_kwDODunzps5oo36i

Fix sliced ConcatenationTable pickling with mixed schemas vertically

closed

MEMBER

A sliced + pickled ConcatenationTable could end up with a different schema than the original schema, if the slice only contains blocks with only a subset of the columns. This can lead to issues when saving datasets from a concatenation of datasets with mixed schemas Reported in https://discuss.huggingface.co/t/datasetdict-save-to-disk-with-num-proc-1-seems-to-hang-with-error/75595

https://api.github.com/repos/huggingface/datasets/issues/6715/timeline

https://api.github.com/repos/huggingface/datasets/issues/6714

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6714/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6714/comments

https://api.github.com/repos/huggingface/datasets/issues/6714/events

https://github.com/huggingface/datasets/pull/6714

PR_kwDODunzps5ooQd2

Expand no-code dataset info with datasets-server info

closed

CONTRIBUTOR

E.g., to have info about a dataset's number of examples for more informative TQDM bars.

https://api.github.com/repos/huggingface/datasets/issues/6714/timeline

https://api.github.com/repos/huggingface/datasets/issues/6713

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6713/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6713/comments

https://api.github.com/repos/huggingface/datasets/issues/6713/events

https://github.com/huggingface/datasets/pull/6713

PR_kwDODunzps5olmqh

Bump huggingface-hub lower version to 0.21.2

closed

MEMBER

This should fix the version compatibility issue when using `huggingface_hub` < 0.21.2 and latest fsspec (>=2023.12.0). See my comment: https://github.com/huggingface/datasets/pull/6687#issuecomment-1976493336 >> EDIT: the fix has been released in `huggingface_hub` 0.21.2 - I removed my commits that were using `huggingface_hub@main` > >Please note that people using `huggingface_hub` < 0.21.2 and latest `fsspec` will have issues when using `datasets`: >- https://github.com/huggingface/lighteval/actions/runs/8139147047/job/22241658122?pr=86 >- https://github.com/huggingface/lighteval/pull/84 CC: @clefourrier

https://api.github.com/repos/huggingface/datasets/issues/6713/timeline

https://api.github.com/repos/huggingface/datasets/issues/6712

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6712/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6712/comments

https://api.github.com/repos/huggingface/datasets/issues/6712/events

https://github.com/huggingface/datasets/pull/6712

PR_kwDODunzps5ok4VF

fix CastError pickling

closed

MEMBER

reported in https://discuss.huggingface.co/t/datasetdict-save-to-disk-with-num-proc-1-seems-to-hang-with-error/75595

https://api.github.com/repos/huggingface/datasets/issues/6712/timeline

https://api.github.com/repos/huggingface/datasets/issues/6711

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6711/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6711/comments

https://api.github.com/repos/huggingface/datasets/issues/6711/events

https://github.com/huggingface/datasets/pull/6711

PR_kwDODunzps5ohM1a

3x Faster Text Preprocessing

open

NONE

I was preparing some datasets for AI training and noticed that `datasets` by HuggingFace uses the conventional `open` mechanism to read the file and split it into chunks. I thought it can be significantly accelerated, and [started with a benchmark](https://gist.github.com/ashvardanian/55c2052e9f78b05b8d614aa90cb12347): ```sh $ pip install --upgrade --force-reinstall datasets $ python benchmark_huggingface_datasets.py xlsum.csv Generating train split: 1004598 examples [00:47, 21116.16 examples/s] Time taken to load the dataset: 48.66838526725769 seconds Time taken to chunk the dataset into parts of size 10000: 0.11466407775878906 seconds Total time taken: 48.78304934501648 seconds ``` For benchmarks I've used a [large CSV file with mixed UTF-8 content](https://github.com/ashvardanian/StringZilla/blob/main/CONTRIBUTING.md#benchmarking-datasets), most common in modern large-scale pre-training pipelines. I've later patched the `datasets` library to use `stringzilla`, which resulted in significantly lower memory consumption and in 2.9x throughput improvement on the AWS `r7iz` instances. That's using slow SSDs mounted over the network. Performance on local SSDs on something like a DGX-H100 should be even higher: ```sh $ pip install -e . $ python benchmark_huggingface_datasets.py xlsum.csv Generating train split: 1004598 examples [00:15, 64529.90 examples/s] Time taken to load the dataset: 16.45028805732727 seconds Time taken to chunk the dataset into parts of size 10000: 0.1291060447692871 seconds Total time taken: 16.579394102096558 seconds ``` I've already [pushed the patches to my fork](https://github.com/ashvardanian/datasets/tree/faster-text-parsers), and would love to contribute them to the upstream repository. --- All the tests pass, but they leave a couple of important questions open. The default Python `open(..., newline=None)` uses universal newlines, where `\n`, `\r`, and `\r\n` are all converted to `\n` on the fly. I am not sure if its a good idea for a general purpose dataset preparation pipeline? I can simulate the same behavior (which I don't yet do) for `"line"` splitter. Adjusting it for `"paragraph"`-splitter would be harder. Should we stick exactly to the old Pythonic behavior or stay closer to how C and other programming languages do that?

https://api.github.com/repos/huggingface/datasets/issues/6711/timeline

https://api.github.com/repos/huggingface/datasets/issues/6710

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6710/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6710/comments

https://api.github.com/repos/huggingface/datasets/issues/6710/events

https://github.com/huggingface/datasets/pull/6710

PR_kwDODunzps5oe4ov

Persist IterableDataset epoch in workers

open

MEMBER

Use shared memory for the IterableDataset epoch. This way calling `ds.set_epoch()` in the main process will update the epoch in the DataLoader workers as well. This is useful especially because the epoch is used to compute the `effective_seed` used for shuffling. I used torch's shared memory in case users want to send dataset copies without shared memory using pickle. I also find it easier to use than using `multiprocessing.shared_memory` than requires unlinking only in the main process, or `mp.Value` that is not picklable. close https://github.com/huggingface/datasets/issues/6673 cc @rwightman

https://api.github.com/repos/huggingface/datasets/issues/6710/timeline

https://api.github.com/repos/huggingface/datasets/issues/6709

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/6709/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/6709/comments

https://api.github.com/repos/huggingface/datasets/issues/6709/events

https://github.com/huggingface/datasets/pull/6709

PR_kwDODunzps5oc2Fg

set dev version

closed

MEMBER

https://api.github.com/repos/huggingface/datasets/issues/6709/timeline