Dataset Features
- Fast streaming decode for image/audio/video/etc. folders datasets by @lhoestq in #8791
- feat(mesh): accept .gltf in the 3D Mesh feature and the meshfolder builder by @li-lizhe in #8958
Other improvements and bug fixes
- Bump doc-builder workflow pin by @paulinebm in #8753
- Make Features.setitem/delitem update only the touched column by @novain-shibu in #8777
- fix(json): defensively recover invalid Arrow tables in JSON loader (#5531) by @FinalSunFlower in #8521
- Close the xopen handle in xpandas_read_csv by @chrikrah in #8786
- bump fsspec and remove aiohttp dep by @lhoestq in #8825
- Fix indices buffer loading in Dataset.from_buffer by @jinchuntan in #8840
- Fix push_to_hub dropping rows when num_proc exceeds num_shards by @behroozazarkhalili in #8795
- Preserve active iterators when reconfiguring threaded decoding by @Linxiushen in #8855
- Reject duplicate WebDataset fields instead of silently overwriting data by @pavanmudumba in #8856
- Fix concatenate_datasets giving every rank all the data after split_dataset_by_node by @hzmsh in #8800
- Unescape doubled quotes in GenBank qualifier values by @chrikrah in #8869
- Fix concatenation cache fingerprints for different axes by @gorkadabo1 in #8865
- Bump doc-builder to get the docs machine-translated by @mishig25 in #8905
- Fix batch(by_column=) default accumulation causing full downloads by @lhoestq in #8960
- Fix batched map on a dataset with an index when the function returns a table by @ViderStar in #8896
- more distributed docs (hsdp) by @lhoestq in #8957
- docs: clarify download modes and transform caching by @original4422 in #8804
- Fix Dataset.take raising IndexError when taking more elements than available by @vidigoat in #8312
- Remove loading of pickled pandas daraframes (CVE-2026-57845) by @SaiTeja-Erukude in #8879
- Fix save_to_disk failing when a batch holds more than 2GB of image or audio files by @hzmsh in #8828
- Fix mixed-type JSON fields with slashes in their names by @its-amann in #8976
- Cast large_string and string_view storage in ClassLabel by @bhaskargurram-ai in #8688
- Preserve headers and document structure when exporting empty datasets by @salma2vec in #8811
New Contributors
- @novain-shibu made their first contribution in #8777
- @FinalSunFlower made their first contribution in #8521
- @chrikrah made their first contribution in #8786
- @jinchuntan made their first contribution in #8840
- @pavanmudumba made their first contribution in #8856
- @hzmsh made their first contribution in #8800
- @gorkadabo1 made their first contribution in #8865
- @li-lizhe made their first contribution in #8958
- @ViderStar made their first contribution in #8896
- @original4422 made their first contribution in #8804
- @vidigoat made their first contribution in #8312
- @SaiTeja-Erukude made their first contribution in #8879
- @its-amann made their first contribution in #8976
- @bhaskargurram-ai made their first contribution in #8688
- @salma2vec made their first contribution in #8811
Full Changelog: 5.1.0...5.2.0