Skip to content

Dataset · likaixin

114k examples for training AI code editing

likaixin/InstructCoder

English dataset of instruction-source-edit triplets for adapting language models to code modification.

What's inside

Contains exactly 114,239 text examples generated using ChatGPT for various source code editing scenarios. Each item is a triplet consisting of an English instruction, an input code snippet, and the final corrected version. The authors note that fine-tuning LLaMA-33B on this data allows it to match stronger models in code editing tasks.

What you can do with it

  • SFT fine-tuning of language models and LoRA for refactoring and code editing
  • Creating IDE AI assistants capable of making targeted changes
  • Testing LLM ability to follow code modification instructions
Tasks
Text generation
Languages
English
Size
100K–1M rows
Downloads this month
20.6K
Likes
18

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("likaixin/InstructCoder")