Parallel corpus of code in 7 languages and descriptions
Code generation benchmark matching English descriptions and seven programming languages.
- Text generation
- Code
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
Code generation benchmark matching English descriptions and seven programming languages.
Largest collection of permissive-licensed Kotlin code from GitHub.
Between 10 and 100 thousand Spanish instruction-response pairs translated using GPT-3.5-turbo for language model fine-tuning.
Between 1 and 10 million semantically equivalent sentences from Tatoeba for 73 languages.
English question-answer pairs based on arXiv scientific papers for training and evaluating QA systems.
A massive corpus of ~30 billion tokens from GitHub repositories on pentesting, OSINT, vulnerability exploitation, and forensics.
Million-scale set of mathematical texts from Tokyo Tech for training generation models.
1,899 tasks where the model must learn new rules and procedures from context rather than pre-training.
FinQA questions with step-by-step GPT-4o reasoning paths for training financial reasoning models.
An English dataset of questions with erroneous assumptions to evaluate LLM pragmatic responses.
A longitudinal benchmark for verifying medication regimen consistency during hospital admissions.
A set of 1k–10k challenging English text tasks where LLMs historically lagged behind humans.
A small document QA benchmark across eight domains for multimodal reasoning.
A corpus with candidate responses from multiple language models, automated metrics, and ChatGPT evaluations for ranking training.
English feedback texts to train open LLMs in fine-grained long answer evaluation.
Around 6k cleaned genre texts and writing instruction prompts in English.
From 10K to 100K safe prompts with provocative wording and reference answers from Amazon Science.
Up to 10k complex English tasks with step-by-step reasoning generated using trace inversion.
An English tabular and text corpus for sentiment classification and translation tasks.
Bilingual corpus of about 40,000 books in Uzbek across Latin and Cyrillic versions.
A collection of research papers, code, chats, and traces from autonomous scientific agents on a preprint server.
Data for training language models in public transit route planning across Chinese cities.
A multi-task English text corpus of over 1 million entries for training small language models.
100k to 1M Japanese instruction-response pairs for SFT generated by Calm3 and Qwen2.5 models.