system HF staff commited on
Commit
643c62a
1 Parent(s): f51786c

Update files from the datasets library (from 1.9.0)

Browse files

Release notes: https://github.com/huggingface/datasets/releases/tag/1.9.0

Files changed (1) hide show
  1. README.md +60 -4
README.md CHANGED
@@ -269,10 +269,17 @@ trivia_qa = load_dataset('trivia_qa', 'unfiltered.nocontext')
269
 
270
  # The KILT IDs can then be mapped to the TriviaQA questions with:
271
  triviaqa_map = {}
 
 
 
 
 
 
 
272
  for k in ['train', 'validation', 'test']:
273
  triviaqa_map = dict([(q_id, i) for i, q_id in enumerate(trivia_qa[k]['question_id'])])
274
  kilt_triviaqa[k] = kilt_triviaqa[k].filter(lambda x: x['id'] in triviaqa_map)
275
- kilt_triviaqa[k] = kilt_triviaqa[k].map(lambda x: {'input': trivia_qa[k][triviaqa_map[x['id']]]['question']})
276
  ```
277
 
278
  ### Supported Tasks and Leaderboards
@@ -406,7 +413,22 @@ Examples from all configurations have the following features:
406
 
407
  ### Data Splits
408
 
409
- [Needs More Information]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
410
 
411
  ## Dataset Creation
412
 
@@ -464,8 +486,42 @@ Examples from all configurations have the following features:
464
 
465
  ### Citation Information
466
 
467
- [Needs More Information]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
468
 
469
  ### Contributions
470
 
471
- Thanks to [@thomwolf](https://github.com/thomwolf), [@yjernite](https://github.com/yjernite) for adding this dataset.
 
269
 
270
  # The KILT IDs can then be mapped to the TriviaQA questions with:
271
  triviaqa_map = {}
272
+
273
+ def add_missing_data(x, trivia_qa_subset, triviaqa_map):
274
+ i = triviaqa_map[x['id']]
275
+ x['input'] = trivia_qa_subset[i]['question']
276
+ x['output']['original_answer'] = trivia_qa_subset[i]['answer']['value']
277
+ return x
278
+
279
  for k in ['train', 'validation', 'test']:
280
  triviaqa_map = dict([(q_id, i) for i, q_id in enumerate(trivia_qa[k]['question_id'])])
281
  kilt_triviaqa[k] = kilt_triviaqa[k].filter(lambda x: x['id'] in triviaqa_map)
282
+ kilt_triviaqa[k] = kilt_triviaqa[k].map(add_missing_data, fn_kwargs=dict(trivia_qa_subset=trivia_qa[k], triviaqa_map=triviaqa_map))
283
  ```
284
 
285
  ### Supported Tasks and Leaderboards
 
413
 
414
  ### Data Splits
415
 
416
+ The configurations have the following splits:
417
+
418
+ | | Train | Validation | Test |
419
+ | ----------- | ----------- | ----------- | ----------- |
420
+ | triviaqa | 61844 | 5359 | 6586 |
421
+ | fever | 104966 | 10444 | 10100 |
422
+ | aidayago2 | 18395 | 4784 | 4463 |
423
+ | wned | | 3396 | 3376 |
424
+ | cweb | | 5599 | 5543 |
425
+ | trex | 2284168 | 5000 | 5000 |
426
+ | structured_zeroshot | 147909 | 3724 | 4966 |
427
+ | nq | 87372 | 2837 | 1444 |
428
+ | hotpotqa | 88869 | 5600 | 5569 |
429
+ | eli5 | 272634 | 1507 | 600 |
430
+ | wow | 94577 | 3058 | 2944 |
431
+
432
 
433
  ## Dataset Creation
434
 
 
486
 
487
  ### Citation Information
488
 
489
+ Cite as:
490
+ ```
491
+ @inproceedings{kilt_tasks,
492
+ author = {Fabio Petroni and
493
+ Aleksandra Piktus and
494
+ Angela Fan and
495
+ Patrick S. H. Lewis and
496
+ Majid Yazdani and
497
+ Nicola De Cao and
498
+ James Thorne and
499
+ Yacine Jernite and
500
+ Vladimir Karpukhin and
501
+ Jean Maillard and
502
+ Vassilis Plachouras and
503
+ Tim Rockt{\"{a}}schel and
504
+ Sebastian Riedel},
505
+ editor = {Kristina Toutanova and
506
+ Anna Rumshisky and
507
+ Luke Zettlemoyer and
508
+ Dilek Hakkani{-}T{\"{u}}r and
509
+ Iz Beltagy and
510
+ Steven Bethard and
511
+ Ryan Cotterell and
512
+ Tanmoy Chakraborty and
513
+ Yichao Zhou},
514
+ title = {{KILT:} a Benchmark for Knowledge Intensive Language Tasks},
515
+ booktitle = {Proceedings of the 2021 Conference of the North American Chapter of
516
+ the Association for Computational Linguistics: Human Language Technologies,
517
+ {NAACL-HLT} 2021, Online, June 6-11, 2021},
518
+ pages = {2523--2544},
519
+ publisher = {Association for Computational Linguistics},
520
+ year = {2021},
521
+ url = {https://www.aclweb.org/anthology/2021.naacl-main.200/}
522
+ }
523
+ ```
524
 
525
  ### Contributions
526
 
527
+ Thanks to [@thomwolf](https://github.com/thomwolf), [@yjernite](https://github.com/yjernite) for adding this dataset.