Text & Language datasets

Text datasets for language model pre-training, instruction tuning, translation and evaluation. Listings range from domain-specific corpora and parallel text to dialogue, question answering and human-written reasoning data.

The catalogue lists 776 text & language datasets. Most listings are text, audio and medical data, typically used for conversational AI, fine-tuning & instruction data and speech recognition. Where providers state a volume in hours, it ranges from 19.5 hours to 25M hours.