hiyouga / LLaMA-Factory

Unify Efficient Fine-Tuning of 100+ LLMs
Apache License 2.0
25.26k stars 3.13k forks source link

LoRA微调和全参微调的时候总是会出现过拟合,在无法提高数据集大小的情况下,应该如何解决这个问题呢 #4452

Closed MemoryOldTime closed 4 days ago

MemoryOldTime commented 4 days ago

Reminder

System Info

8张Ascend910A,在modelarts上运行代码

Reproduction

bash examples/train_full/train.sh

Expected behavior

我想要在某一专业领域实现微调,希望能够解决过拟合的问题

Others

hiyouga commented 4 days ago

混合通用领域数据

MemoryOldTime commented 4 days ago

混合通用领域数据

625 但这eval loss 不断从来没有下来过