Legal tasks benchmark for evaluating LLMs
A collection of English legal test tasks for evaluating model domain adaptation.
- Text classification
- Question answering
- Zero shot classification
- English
Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,788 in the catalog so far, with new ones added every hour.
Answered by Gemini 4 · the chat is kept only in this tab
A collection of English legal test tasks for evaluating model domain adaptation.
A synthetic dataset of 2.5 million clinical instructions in English for SFT and DPO alignment.
100K-1M textbook-style English texts for training small language models.
Between 100 million and 1 billion Russian customer reviews from the Wildberries marketplace in zstd archives with jsonl.
Digitized historical corpus of Greek parliamentary debates with links to primary sources.
English math preference dataset in DPO format; card is nearly empty.
19,331 multi-step reasoning dialogues in ChatML + Hermes format in English for training local agent models.
801 thousand synthetic instruction-response pairs for SFT training of programming models in Japanese and English.
English multi-turn dialogues for LLM alignment and preference training (RLHF/DPO).
An archive of all questions and answers from the Law section of StackExchange in English up to August 2023.
A set of synthetic prompts and haiku poems for training language models to follow genre structure.
1k to 10k task execution logs of the GLM-5 model using terminal, code, and browser tools.
A small collection of everyday queries where popular AI assistants made notable errors.
From 10 to 100 thousand high-quality Wikipedia articles with 179 million tokens preserving markup for generation and summarization.
Bhagavad Gita verses in Sanskrit with Hindi and English translations for translation and generation tasks.
Small set of tabular and JSON metric reports for reproducible experiment analysis.
A corpus of English web texts (1-10M entries) for training models in objective analysis and bias neutralization.
6,105 posts and 124 communities from a social network for AI agents, collected before an influx of humans and third-party bots.
207 thousand code examples in various languages with detailed explanations in a mini-textbook format.
Over 270k web documents (~230M tokens) in English, Yoruba, Hausa, and Igbo.
10,000 to 100,000 recipes with titles, ingredient lists, and step-by-step instructions for fine-tuning language models.
Augmented English web texts detailing theoretical concepts and practical examples.
Filtered short texts from RefinedWeb and MiniPile for training compact language models.