huggingface/datasets 2.16.0

on GitHub

Security features

• Add trust_remote_code argument by @lhoestq in #6429
  • Some Hugging Face datasets contain custom code which must be executed to correctly load the dataset. The code can be inspected in the repository content at https://hf.co/datasets/<repo_id>. A warning is shown to let the user know about the custom code, and they can avoid this message in future by passing the argument trust_remote_code=True.
  • Passing trust_remote_code=True will be mandatory to load these datasets from the next major release of datasets.
  • Using the environment variable HF_DATASETS_TRUST_REMOTE_CODE=0 you can already disable custom code by default without waiting for the next release of datasets
• Use parquet export if possible by @lhoestq in #6448
  • This allows loading most old datasets based on custom code by downloading the Parquet export provided by Hugging Face
  • You can see a dataset's Parquet export at https://hf.co/datasets/<repo_id>/tree/refs%2Fconvert%2Fparquet

Features

• Webdataset dataset builder by @lhoestq in #6391
• Implement get dataset default config name by @albertvillanova in #6511
• Lazy data files resolution and offline cache reload by @lhoestq in #6493
  • This speeds up the load_dataset step that lists the data files of big repositories (up to x100) but requires huggingface_hub 0.20 or newer
  • Fix load_dataset that used to reload data from cache even if the dataset was updated on Hugging Face
  • Reload a dataset from your cache even if you don't have internet connection
  • New cache directory scheme for no-script datasets: ~/.cache/huggingface/datasets/username___dataset_name/config_name/version/commit_sha
  • Backward comaptibility: cached datasets from datasets 2.15 (using the old scheme) are still reloaded from cache

General improvements and bug fixes

• Remove unused argument in _get_data_files_patterns by @lhoestq in #6343
• Set usedforsecurity=False in hashlib methods (FIPS compliance) by @Wauplin in #6414
• Use ruff for formatting by @mariosasko in #6434
• Create DatasetNotFoundError and DataFilesNotFoundError by @albertvillanova in #6431
• Fix multi gpu map example by @lhoestq in #6415
• Better tqdm wrapper by @mariosasko in #6433
• Remove Table.__getstate__ and Table.__setstate__ by @LZHgrla in #6444
• Use filelock package for file locking by @mariosasko in #6445
• Fix metadata file resolution when inferred pattern is ** by @mariosasko in #6449
• Update hub-docs reference by @mishig25 in #6453
• Refactor dill logic by @mariosasko in #6454
• Don't require trust_remote_code in inspect_dataset by @lhoestq in #6456
• [docs] troubleshooting guide by @MKhalusova in #6424
• Missing DatasetNotFoundError by @lhoestq in #6462
• Disable benchmarks in PRs by @lhoestq in #6463
• More robust temporary directory deletion by @mariosasko in #6426
• Fix shard retry mechanism in push_to_hub by @mariosasko in #6461
• Use auth to get parquet export by @lhoestq in #6468
• Remove delete doc CI by @lhoestq in #6471
• Fix CI quality by @albertvillanova in #6473
• Fix PermissionError on Windows CI by @albertvillanova in #6477
• More robust preupload retry mechanism by @mariosasko in #6479
• Add IterableDataset __repr__ by @lhoestq in #6480
• Fix max lock length on unix by @lhoestq in #6482
• Fix ArrayXD YAML conversion by @mariosasko in #6168
• Fix docs phrasing about supported formats when sharing a dataset by @albertvillanova in #6486
• Fix deprecation warning when building conda package by @albertvillanova in #6425
• Make push_to_hub return CommitInfo by @albertvillanova in #6492
• docs: add reference Git over SSH by @severo in #6499
• Fallback on dataset script if user wants to load default config by @lhoestq in #6498
• Don't expand_info in HF glob by @lhoestq in #6469
• Fix streaming xnli by @lhoestq in #6503
• Pickle support for torch.Generator objects by @mariosasko in #6502
• Enable setting config as default when push_to_hub by @albertvillanova in #6500
• Better cast error when generating dataset by @lhoestq in #6509
• Replace list_files_info with list_repo_tree in push_to_hub by @mariosasko in #6510
• Remove deprecated HfFolder by @lhoestq in #6512
• Support huggingface-hub pre-releases by @albertvillanova in #6516
• Support push_to_hub canonical datasets by @albertvillanova in #6519
• Support commit_description parameter in push_to_hub by @albertvillanova in #6520
• fix get_metadata_patterns function args error by @d710055071 in #6518
• Fix metrics dead link by @qgallouedec in #6491
• fix tests by @lhoestq in #6523
• Cache backward compatibility with 2.15.0 by @lhoestq in #6514
• Preserve order of configs and splits when using Parquet exports by @albertvillanova in #6526

New Contributors

• @LZHgrla made their first contribution in #6444
• @d710055071 made their first contribution in #6518

Full Changelog: 2.15.0...2.16.0