{"url":"/dataset/swag","name":"SWAG","full_name":"Situations With Adversarial Generations","description_markdown":"Given a partial description like \"she opened the hood of the car,\" humans can reason about the situation and anticipate what might come next (\"then, she examined the engine\"). SWAG (Situations With Adversarial Generations) is a large-scale dataset for this task of grounded commonsense inference, unifying natural language inference and physically grounded reasoning.\r\n\r\nThe dataset consists of 113k multiple choice questions about grounded situations. Each question is a video caption from LSMDC or ActivityNet Captions, with four answer choices about what might happen next in the scene. The correct answer is the (real) video caption for the next event in the video; the three incorrect answers are adversarially generated and human verified, so as to fool machines but not humans. The authors aim for SWAG to be a benchmark for evaluating grounded commonsense NLI and for learning representations.\r\n\r\nSource: [SWAG](https://rowanzellers.com/swag/)\r\nImage Source: [Zellers et al](https://arxiv.org/pdf/1808.05326v1.pdf)","description_withheld":null,"homepage":"https://rowanzellers.com/swag/","introduced_date":"2018-08-16","introduced_date_note":null,"introduced_by":{"paper":"/paper/swag-a-large-scale-adversarial-dataset-for","title":"SWAG: A Large-Scale Adversarial Dataset for Grounded Commonsense Inference","first_author":"Rowan Zellers","url":null},"license":{"name":"MIT","url":"https://github.com/rowanz/swagaf/blob/master/LICENSE"},"modalities":[{"name":"Texts","url":"/datasets/modality/texts"}],"tasks":[{"name":"Question Answering","url":"/task/question-answering","datasets_with_task":"/datasets/task/question-answering"},{"name":"Common Sense Reasoning","url":"/task/common-sense-reasoning","datasets_with_task":"/datasets/task/common-sense-reasoning"}],"languages":[{"name":"English","url":"/datasets/language/english"}],"variants":["SWAG"],"data_loaders":[{"repo":"https://github.com/huggingface/datasets","url":"https://huggingface.co/datasets/allenai/swag","frameworks":["tf","pytorch","jax"]},{"repo":"https://github.com/huggingface/datasets","url":"https://huggingface.co/datasets/swag","frameworks":["tf","pytorch","jax"]},{"repo":"https://github.com/activeloopai/Hub","url":"https://docs.activeloop.ai/datasets/swag-dataset","frameworks":["tf","pytorch"]},{"repo":"https://github.com/allenai/allennlp-models","url":"https://docs.allennlp.org/models/main/models/mc/dataset_readers/swag/","frameworks":["pytorch"]}],"num_papers_in_archive":163,"source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28"},"benchmarks":[{"leaderboard":"/sota/common-sense-reasoning-on-swag","task":"Common Sense Reasoning","dataset_variant":"SWAG","rows":5,"metrics":["Test","Dev"],"first_row_in_archive_order":{"model":"DeBERTalarge","paper":"/paper/deberta-decoding-enhanced-bert-with","metrics":{"Test":"90.8"},"code_links":[{"title":"huggingface/transformers","url":"https://github.com/huggingface/transformers"},{"title":"microsoft/DeBERTa","url":"https://github.com/microsoft/DeBERTa"},{"title":"osu-nlp-group/mind2web","url":"https://github.com/osu-nlp-group/mind2web"},{"title":"neuralmind-ai/coliee","url":"https://github.com/neuralmind-ai/coliee"},{"title":"huberemanuel/DeBERTa","url":"https://github.com/huberemanuel/DeBERTa"},{"title":"Jason-J-Choi/DeBERTa_TxtClassifier","url":"https://github.com/Jason-J-Choi/DeBERTa_TxtClassifier"},{"title":"2024-MindSpore-1/Code2","url":"https://github.com/2024-MindSpore-1/Code2/tree/main/model-1/deberta"},{"title":"MindCode-4/code-3","url":"https://github.com/MindCode-4/code-3/tree/main/deberta"},{"title":"2024-MindSpore-1/Code2","url":"https://github.com/2024-MindSpore-1/Code2/tree/main/model-1/deberta_v2"},{"title":"MindCode-4/code-3","url":"https://github.com/MindCode-4/code-3/tree/main/deberta_v2"},{"title":"JONGWE1/Paddle-DeBERTa","url":"https://github.com/JONGWE1/Paddle-DeBERTa"},{"title":"Mind23-2/MindCode-161","url":"https://github.com/Mind23-2/MindCode-161"},{"title":"pwc-1/Paper-8","url":"https://github.com/pwc-1/Paper-8/tree/main/deberta"},{"title":"pwc-1/Paper-8","url":"https://github.com/pwc-1/Paper-8/tree/main/deberta_v2"}]},"note":"rows are the archive's own order at snapshot; nothing here re-ranks them"},{"leaderboard":"/sota/question-answering-on-swag","task":"Question Answering","dataset_variant":"SWAG","rows":1,"metrics":["Accuracy"],"first_row_in_archive_order":{"model":"DeBERTaV3large","paper":"/paper/debertav3-improving-deberta-using-electra","metrics":{"Accuracy":"93.4"},"code_links":[{"title":"microsoft/DeBERTa","url":"https://github.com/microsoft/DeBERTa"},{"title":"dashenzi721/hra","url":"https://github.com/dashenzi721/hra"},{"title":"stareru/csqa_debertav3","url":"https://github.com/stareru/csqa_debertav3"}]},"note":"rows are the archive's own order at snapshot; nothing here re-ranks them"}],"papers_with_a_benchmark_row":[{"paper":"/paper/debertav3-improving-deberta-using-electra","title":"DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing","date":"2021-11-18","rows_on_this_dataset":1,"code_links":3,"syntology":{"read_at":"2026-09-24T18:15:14+00:00","samples_harvested":7,"samples_ran":0,"samples_unverified":7,"pointer_only_for_licence":0,"claim":"Per-sample execution on synthesized fixtures; not a correctness claim."}},{"paper":"/paper/deberta-decoding-enhanced-bert-with","title":"DeBERTa: Decoding-enhanced BERT with Disentangled Attention","date":"2020-06-05","rows_on_this_dataset":1,"code_links":14,"syntology":{"read_at":"2026-09-24T18:15:14+00:00","samples_harvested":13,"samples_ran":4,"samples_unverified":9,"pointer_only_for_licence":3,"claim":"Per-sample execution on synthesized fixtures; not a correctness claim."}},{"paper":"/paper/roberta-a-robustly-optimized-bert-pretraining","title":"RoBERTa: A Robustly Optimized BERT Pretraining Approach","date":"2019-07-26","rows_on_this_dataset":1,"code_links":67,"syntology":{"read_at":"2026-09-24T18:15:14+00:00","samples_harvested":48,"samples_ran":22,"samples_unverified":26,"pointer_only_for_licence":23,"claim":"Per-sample execution on synthesized fixtures; not a correctness claim."}},{"paper":"/paper/bert-pre-training-of-deep-bidirectional","title":"BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding","date":"2018-10-11","rows_on_this_dataset":1,"code_links":534,"syntology":{"read_at":"2026-09-24T18:15:14+00:00","samples_harvested":659,"samples_ran":204,"samples_unverified":455,"pointer_only_for_licence":149,"claim":"Per-sample execution on synthesized fixtures; not a correctness claim."}},{"paper":"/paper/swag-a-large-scale-adversarial-dataset-for","title":"SWAG: A Large-Scale Adversarial Dataset for Grounded Commonsense Inference","date":"2018-08-16","rows_on_this_dataset":2,"code_links":0,"syntology":null}],"syntology_totals":{"read_at":"2026-09-24T18:15:14+00:00","papers_with_samples":4,"samples_harvested":727,"samples_ran":230,"samples_unverified":497,"pointer_only_for_licence":175,"papers_with_no_sample_that_ran":1,"note":"the per-paper counts above, summed; not a rate"},"papers_note":"The archive never published its papers-using-dataset list; these are papers with a leaderboard row on this dataset's benchmarks."}