Llama3.1 8B 使用《史记》七十列传文本数据微调训练,实现现代文翻译至古文,效果还不错! | colab | unsloth | hugging face | 大模型微调
训练平台Colab: https://colab.research.google.com/drive/1Ne068_p8ZbJt93DdXT2zlxcTIYUuE_zT?usp=sharing
微调工具: https://github.com/unslothai/unslothhttps://github.com/ggerganov/llama.cpp
文言文(古文)- 现代文平行语料: https://github.com/NiuTrans/Classical-Modern生成训练数据集时所使用的convert.py: https://gist.github.com/lanesky/6092906644c36d16ad39df3ac6d623d2
提交到Huggingface上的训练数据集: https://huggingface.co/datasets/AISPIN/shiji-70liezhuan
Push到Huggingface上的训练好的Model: https://huggingface.co/AISPIN/Llama-3.1-8B-bnb-4bit-wenyanwen
LM Studio下载地址: https://lmstudio.ai/