SKU RQA-5007 · Sold by External
DROP
Product specifications
| SKU | RQA-5007 |
|---|---|
| Data type | Reasoning QA |
| Volume | 86,935 questions (77,400 train + 9,535 validation) |
| Size on disk | ~117 MB uncompressed (dataset_size 117,310,261 bytes); ~11.5 MB Parquet download (download_size 11,538,387 bytes) |
| Format | Parquet (Hugging Face); originally distributed as JSON |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open dataset (HF) |
| Quality score | — |
| License | CC-BY-SA-4.0 |
DROP (Discrete Reasoning Over Paragraphs) is a reading-comprehension QA benchmark built by UCI NLP and AI2 in which answering each question requires performing discrete operations such as addition, subtraction, counting, sorting, and comparison over a passage. The public Hugging Face release (ucinlp/drop, config 'default') contains 77,400 training and 9,535 validation examples; the original benchmark reports 96,567 questions including a held-out test set not distributed here. Each record pairs a passage with a question and a set of gold answer spans, where each answer carries a type label such as number, span, or date. Answers may be a number, a date, or one or more text spans extracted from the passage. The official metric is a numerically-aware Exact Match plus macro-averaged F1 computed against the annotated gold answer set. The dataset is released under CC-BY-SA-4.0.