Dataset · m-a-p
High-quality Chinese instructions for SFT
m-a-p/COIG-CQIA
A high-quality Chinese instruction dataset for SFT sourced from web Q&A and articles.
What's inside
The dataset contains instructions and responses in Chinese; according to the card, it was sourced from Q&A and articles on the Chinese web. The data underwent deep cleaning, rewriting, and manual verification, with a size of 10K to 100K examples. Format is text, and tasks include question-answering, text-generation, and text-classification.
What you can do with it
- Fine-tuning a Chinese instruct model (SFT) on a small, high-quality dataset
- Creating a LoRA for a Chinese assistant
- Testing a model on instruction following and human-like dialogue responses
- Tasks
- Question answering, Text classification, Text generation
- Languages
- Chinese
- Size
- 10K–100K rows
- Downloads this month
- 11.9K
- Likes
- 796
Models trained on it
- TiaoFeng/Spark-X2.6-4B-EAGLE3-Chinese-retrain 765
- UnderTides/DeepSeek-V4.1-Flash-1017-heretic-loraImage text → text 147
- UnderTides/DeepSeek-V4.1-Flash-1017-heretic-lora-devImage text → text 102
- happyme531/Granite-4.2-3B-W4A8-Mixed-QQQ-SQ-RK3588Text generation 94
- happyme531/MiniCPM5-2B-W4A8-Mixed-QQQ-Rot-RK3588Text generation 72
- TJUNLP/FuxiTranyu-8B-SFTText generation 46
- yiwenX/RiverValley-1B-Experimental-20261110Text generation 32
- xiaozhengzhong/Qwen3-8B-Tieba-Magic-LoRAText generation 29
How to load it
from datasets import load_dataset
ds = load_dataset("m-a-p/COIG-CQIA")