Dataset · google-research-datasets
Complex Sentence Pairs for Paraphrase Identification (PAWS)
google-research-datasets/paws
Over 760K English sentence pairs with similar words but different meanings to test context understanding.
What's inside
The English-language dataset contains 108,463 human-labeled pairs and 656,000 automatically labeled pairs. Examples are gathered from Wikipedia and the Quora question database with word permutations to test how well a model captures sentence structure and context rather than mere lexical overlap.
What you can do with it
- Training classifiers to determine semantic equivalence and duplicates
- Testing the robustness of text embeddings to word permutations
- Fine-tuning NLI and semantic search models on complex examples
- Tasks
- Text classification
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 51.5K
- Likes
- 54
Models trained on it
- StrandsAgents/strands-decider-2B-qwen3.5-v1-2610Text classification 25K
- StrandsAgents/strands-decider-2B-hobson-v21Text classification 15.3K
- StrandsAgents/strands-decider-2B-hobson-v19Text classification 7K
- tasksource/ettin-17m-embedSentence similarity 5.6K
- StrandsAgents/strands-decider-E4B-gemma4-v1-2610Text classification 1.4K
- StrandsAgents/strands-decider-12B-gemma4-v1-2610Text classification 1.1K
- manjunathshiva/opendecider-nanoZero shot classification 1.1K
- mradermacher/elythros-decider-1.7b-GGUFZero shot classification 887
How to load it
from datasets import load_dataset
ds = load_dataset("google-research-datasets/paws")