SKU SFT-2404 · Sold by External

No Robots

Product specifications

SKUSFT-2404
Data typeSFT dataset
Volume10,000 instruction-demonstration pairs
Size on disk22.1 MB
Formatparquet
Access modelPUBLIC LICENSE
PricingFree · open dataset (HF)
Quality score
LicenseCreative Commons Attribution Non Commercial 4.0 (CC BY-NC 4.0)
No Robots is a supervised fine-tuning (SFT) dataset of 10,000 instructions with paired demonstrations, all created by skilled human annotators rather than generated by a language model. It was released by the Hugging Face H4 team and is modeled after the SFT data described in OpenAI's InstructGPT paper. Each record contains a user prompt, a SHA-256 prompt_id, a messages list holding the user turn and the gold human-written assistant response, and one of 10 category labels (Generation, Open QA, Brainstorm, Chat, Rewrite, Summarize, Coding, Classify, Closed QA, Extract). The data is split into 9,500 train and 500 test examples and is distributed under the CC BY-NC 4.0 (non-commercial) license. It is mostly single-turn human demonstration data (the Chat category can include a system turn and multiple turns), and there is no automated grader: the human-authored assistant message serves as the reference target for imitation.