Skip to content

Dataset · HuggingFaceH4

10k Human Instructions for SFT

HuggingFaceH4/no_robots

10k instructions and human responses for SFT to help models follow commands better.

What's inside

The dataset contains 10k instructions and demonstrations created manually by qualified annotators, without GPT generation. It is modeled after the instruction dataset description from the OpenAI InstructGPT paper and mostly consists of single-turn examples. The language is English, the size ranges from 10k to 100k records, and the data is suitable for supervised fine-tuning (SFT).

What you can do with it

  • Fine-tune a model to follow instructions via SFT
  • Create LoRA for instruction following
  • Compare human and synthetic instructions
Tasks
Text generation
Languages
English
Size
10K–100K rows
Downloads this month
54.9K
Likes
609

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("HuggingFaceH4/no_robots")