Issues

concatenate_datasets drops the split_dataset_by_node split, so every rank gets all the data

huggingface/datasets#8799 · opened

on Oct 14, 2026

1

Fatal `PyGILState_Release` at interpreter exit when streaming Parquet with pyarrow 25 (arrow#45214 workaround is gated to pyarrow <= 24)

huggingface/datasets#8793 · opened

on Oct 13, 2026

1

re-pushing all dataset partitions corrupts it

huggingface/datasets#8789 · opened

on Oct 12, 2026

1

Dataset.add_item fails when an existing struct-backed feature meets an inferred dictionary

huggingface/datasets#8766 · opened

on Oct 7, 2026

1

Dataset.shard raises IndexError for the empty shards when num_shards is larger than the dataset

huggingface/datasets#8763 · opened

on Oct 7, 2026

1

Audio array remains NumPy after Torch formatting

huggingface/datasets#8755 · opened

on Oct 6, 2026

1

Parquet export does not close its writer after a batch or metadata failure

huggingface/datasets#8743 · opened

on Oct 5, 2026

Parquet file with an empty row group fails to load: `ArrowInvalid: BatchSize must be greater than 0`

huggingface/datasets#8740 · opened

on Oct 5, 2026

1

concatenate_datasets(axis=1) crashes on arrow-backed IterableDatasets with different lengths

huggingface/datasets#8719 · opened

on Oct 1, 2026

1

Casting mixed JSON and plain strings to Json depends on row order and batch size

huggingface/datasets#8714 · opened

on Sep 30, 2026

1

JSON loader turns strings like "123" or "null" into other types in inferred Json columns

huggingface/datasets#8713 · opened

on Sep 30, 2026

1

string_to_dict maps fields to unrelated regex capture groups

huggingface/datasets#8709 · opened

on Sep 29, 2026

1

Skip constructing disabled `hf_tqdm` progress bars in `map_nested` and `_single_map_nested`

huggingface/datasets#8705 · opened

on Sep 28, 2026

Avoid repeated `np.lib.NumpyVersion(np.version)` parsing per column/row in `NumpyArrowExtractor._arrow_array_to_numpy`

huggingface/datasets#8704 · opened

on Sep 28, 2026

Windows: `load_dataset` fails with WinError 206 for moderately deep `cache_dir` because the lock filename duplicates the whole path

huggingface/datasets#8702 · opened

on Sep 27, 2026

Casting `large_string` / `string_view` to ClassLabel fails, so `Dataset.from_pandas(df, features=ClassLabel)` breaks with pandas 3

huggingface/datasets#8681 · opened

on Sep 24, 2026

load_dataset("json") fails on mixed-shape JSONL when a chunksize boundary separates the shapes

huggingface/datasets#8675 · opened

on Sep 23, 2026

1

TranslationVariableLanguages crashes on empty translations with declared languages

huggingface/datasets#8659 · opened

on Sep 21, 2026

Long lock filenames do not provide mutual exclusion across Python processes

huggingface/datasets#8653 · opened

on Sep 21, 2026

FileLock wrapper drops timeout and every other constructor argument

huggingface/datasets#8637 · opened

on Sep 17, 2026

2

Probabilistic interleave without replacement samples empty datasets

huggingface/datasets#8626 · opened

on Sep 15, 2026

`DataFilesList.filter()` leaves `origin_metadata` misaligned with the filtered files

huggingface/datasets#8618 · opened

on Sep 14, 2026

1

`SplitDict.copy()` crashes when `dataset_name` is not set

huggingface/datasets#8617 · opened