2.2.0

Dataset Changes

• New: ImageNet by @apsdehal in #4178
  • Manual download only for now
• New: Google Conceptual Captions by @abhishekkrthakur in #1459
• New: Conceptual 12M by @thomasw21 in #4162
• New: Visual Genome by @thomasw21 in #4161
• New: RVL-CDIP by @dnaveenr in #4050
• New: Text-based NP Enrichment (TNE) by @yanaiela in #4153
• New: TextVQA by @apsdehal in #3967
• New: ETT time series dataset by @kashif in #4213
• Update: assin2 - update metadata by @lhoestq in #4172
• Update: Librispeech - Add 'all' config by @patrickvonplaten in #4184
• Update: XGLUE - Support streaming dataset by @albertvillanova in #4249
• Update: crd3 - group all the turns in one example by @shanyas10 in #4240
• Update: pubmed_qa - Remove google drive URL by @lhoestq in #4255
• Update: SAMSum - Replace data URL dataset and support streaming by @albertvillanova in #4254
• Update: SAMSum - Replace data URL dataset within the same repository by @albertvillanova in #4267
• Update: big_patent - Replace data URL in dataset and support streaming by @albertvillanova in #4236
• Update: openbookqa - Add missing features for additional config by @albertvillanova in #4278
• Update: commonsense_qa - Add missing features by @albertvillanova in #4280
• Fix: Common Voice - Make sure bytes are correctly deleted if path exists by @patrickvonplaten in #4212
• Fix: openbookqa - fix bug in choices labels by @manandey in #4259
• Fix: openbookqa - fix style in openbookqa dataset by @albertvillanova in #4270

Dataset Features

• Add support for metadata files to imagefolder by @mariosasko in #4069
  • load a folder of images and metadata stored in metadata.jsonl, more info in the documentation on how to load an image dataset
• Infer splits from the data_dir parameter when loading datasets without script by @polinaeterna in #4144
  • splits are inferred from the directory and file names, see more info in the documentation on how to structure your repository
• Enable label alignment for token classification datasets by @lewtun in #4277
• Add drop_last_batch to IterableDataset.map by @mariosasko in #4215
• Load dataset with TSV files by @albertvillanova in #4246

Dataset Cards

• Autoeval config by @NRajani in #4234
  • Add train-deval-index metadata to automate evaluation on your datasets based on their tasks
• Adding license information for Openbookcorpus by @meg-huggingface in #3525
• Make code for image downloading from image urls cacheable by @mariosasko in #4218
• Fix description links in dataset cards by @albertvillanova in #4222
• Add YAML tags to Dataset Card rotten tomatoes by @mo6zes in #4262
• Remove a copy-paste sentence in dataset cards by @albertvillanova in #4281
• Update LexGLUE README.md by @iliaschalkidis in #4285
• leadboard info added for TNE by @yanaiela in #4273
• Add Lahnda language tag by @mariosasko in #4286
• Add license and point of contact to big_patent dataset by @albertvillanova in #4269
• Add HF Speech Bench to Librispeech Dataset Card by @sanchit-gandhi in #4266

Metrics Changes

• Perplexity Speedup by @emibaylor in #4108
• Add AUC ROC Metric by @emibaylor in #4158
• Small fixes in ROC AUC docs by @wschella in #4239
• Fix/start token mask issue and update documentation by @TristanThrush in #4258
• Add pearsonr mc, update functionality to match the original docs by @emibaylor in #4226

Metric Cards

• Metric card for the XTREME-S dataset by @sashavor in #4251
• Creating metric card for MAE by @sashavor in #4252
• Create metric cards for mean IOU by @sashavor in #4253
• Create metric card for Mahalanobis Distance by @sashavor in #4257
• Create metric card for MSE by @sashavor in #4256
• Fix exact match by @emibaylor in #4166
• Fix google bleu typos, examples by @emibaylor in #4165
• Add f1 metric card, update docstring in py file by @emibaylor in #4227
• Add Recall Metric Card by @emibaylor in #4204
• Matthews Correlation Metric Card by @emibaylor in #4110
• Add Precision Metric Card by @emibaylor in #4203
• Add Accuracy Metric Card by @emibaylor in #4223
• Add Spearmanr Metric Card by @emibaylor in #4109
• Metric card template by @emibaylor in #3915

Documentation

• Document save_to_disk and push_to_hub on images and audio files by @lhoestq in #4193
• Add to docs how to load from local script by @albertvillanova in #4200
• Add code examples to API docs by @stevhliu in #4168
• Add code examples for DatasetDict by @stevhliu in #4245
• Add API code examples for IterableDataset by @stevhliu in #4274
• Add packaged builder configs to the documentation by @lhoestq in #4307
• [Imagefolder] Docs + Don't infer labels from file names when there are metadata + Error messages when metadata and images aren't linked correctly by @lhoestq in #4311

General improvements and bug fixes

• Generate tasks.json taxonomy from huggingface_hub by @julien-c in #4154
• Fix when map function modifies input in-place by @thomasw21 in #4174
• Support streaming cnn_dailymail dataset by @albertvillanova in #4188
• Don't duplicate data when encoding audio or image by @lhoestq in #4187
• Fix outdated docstring about default dataset config by @lhoestq in #4186
• Deprecate shard_size in push_to_hub in favor of max_shard_size by @mariosasko in #4190
• Fix some type annotation in doc by @thomasw21 in #4202
• Update GH template for dataset viewer issues by @albertvillanova in #4201
• Update auth when mirroring datasets on the hub by @lhoestq in #4242
• Rename imagenet2012 -> imagenet-1k by @lhoestq in #4263
• Skip checksum computation in Imagefolder by default by @mariosasko in #4214
• Fix convert_file_size_to_int for kilobits and megabits by @mariosasko in #4205
• Fix typo in logging docs by @stevhliu in #4272
• Bump PyArrow Version to 6 by @dnaveenr in #4250
• task id update by @NRajani in #4244
• Avoid recursion error in map if example is returned as dict value by @mariosasko in #4216
• Update minimal PyArrow version warning by @mariosasko in #4279
• [Minor edit] Fix typo in class name by @cakiki in #4207
• Stream private zipped images by @lhoestq in #4173
• Fix filesystem docstring by @stevhliu in #4283
• Document how to use FAISS index for special operations by @albertvillanova in #4189
• Contributing MedMCQA dataset by @monk1337 in #4064
• Don't do unnecessary list type casting to avoid replacing None values by empty lists by @lhoestq in #4282
• Fix missing lz4 dependency for tests by @albertvillanova in #4295
• Altered faiss installation comment by @vishalsrao in #4220
• Fix CLI run_beam save_infos by @albertvillanova in #4294
• Add missing faiss import to fix #4287 by @alvarobartt in #4288

New Contributors

• @shanyas10 made their first contribution in #4240
• @apsdehal made their first contribution in #4178
• @wschella made their first contribution in #4239
• @TristanThrush made their first contribution in #4258
• @yanaiela made their first contribution in #4153
• @mo6zes made their first contribution in #4262
• @NRajani made their first contribution in #4244
• @sanchit-gandhi made their first contribution in #4266
• @cakiki made their first contribution in #4207
• @monk1337 made their first contribution in #4064
• @alvarobartt made their first contribution in #4288

Full Changelog: 2.1.0...2.2.0