LEXam legal reasoning benchmark
340 legal exams from Switzerland, the EU, and international law for evaluating reasoning models.
- Text classification
- Text generation
- English
- German
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
340 legal exams from Switzerland, the EU, and international law for evaluating reasoning models.
A set of 1,065 textual descriptions of patient complaints labeled with 22 diagnoses for training classifiers.
Over 150 million bibliographic records from the Open Library project for NLP, search, and data extraction tasks.
112,570 English Danbooru tags annotated with 7 categories and 33 slots for prompt analysis and generation.
A small set of ChatGPT jailbreak prompts for studying constraint bypassing.
Hundreds of thousands of texts in six European languages for training models to hide personal information.
English STS Benchmark and its translations into five languages for training sentence embeddings.
Large English, Chinese, and Arabic corpus with manual syntactic and semantic annotation.
Dataset for binary classification of source code for vulnerabilities.
English email dataset for training and testing phishing filters.
A collection of 21,318 prompts and attacks (jailbreaks, toxicity) for testing and improving neural network safety.
Combined dialogue data for QA, summarization, generation, classification, and table-to-text.
Set of tasks for training and evaluating NLP models in Indonesian.
An English-language dataset of 740,957 examples for distilling compact classifiers that evaluate states and decisions in a single pass.
Text collection in 6 languages annotated with positive, negative, and neutral labels for training emotion classifiers.
Over 1.5 million news article and summary pairs in Russian, German, Spanish, French, and Turkish.
A set of 3,048 similar and dissimilar medical question pairs manually authored and labeled by physicians.
Cleaned texts of 342,782 Wikipedia articles with three-level category labeling.
Large fiction text corpus in Japanese for generation and classification.
Large structured archive of Indian judicial acts, tribunal decisions, and legislation in English.
A dataset with real products, images, and descriptions for category classification.
363k unique instruction examples without RLHF, cleaned of duplicates for training generative models.
System prompts and user attacks for training LLM defense against hacking and injections.
Over a million Vietnamese legal documents from the official Ministry of Justice portal, including HTML texts and a link graph between acts.