hiyouga LLaMA-Factory Q-a Discussions
Pinned Discussions
Q&A Discussions
Ask the community for help
-
You must be logged in to vote -
You must be logged in to vote -
You must be logged in to vote Full-Finetune worse than Lora-Finetune 全量fientuee效果差于lora
pendingThis problem is yet to be addressed -
You must be logged in to vote -
You must be logged in to vote -
You must be logged in to vote -
You must be logged in to vote 二次预训练阶段全参微调,损失曲线是否正常，如何优化
pendingThis problem is yet to be addressed -
You must be logged in to vote lora finutune QWen2.5 0.5B use CPU lora微调QWen2.5 0.5B训练使用了cpu，怎么修改为使用cuda
wontfixThis will not be worked on -
You must be logged in to vote -
You must be logged in to vote 使用llama3_lora_pretrain.yaml微调模型后，模型生成能力严重丧失
pendingThis problem is yet to be addressed -
You must be logged in to vote 训练集output长度1万8，微调时cutoff_len设置为2万，效果不好怎么办？
pendingThis problem is yet to be addressed -
You must be logged in to vote -
You must be logged in to vote 0.9.0版本的label构建方式
pendingThis problem is yet to be addressed -
You must be logged in to vote -
You must be logged in to vote What is the correct meaning of the cutoff_len parameter?
pendingThis problem is yet to be addressed -
You must be logged in to vote -
You must be logged in to vote 微调过后想要进行评估，该如何评估，计算blue和rouge分布，数据该如何自定义
pendingThis problem is yet to be addressed -
You must be logged in to vote 请教多轮工具调用的数据格式
pendingThis problem is yet to be addressed -
You must be logged in to vote -
You must be logged in to vote -
You must be logged in to vote 微调qwen2-vl让他识别化学结构输出smiles码但是结果不好
pendingThis problem is yet to be addressed -
You must be logged in to vote -
You must be logged in to vote 全量微调后的文件怎么这大，还有这个checkpoint文件是什么？
pendingThis problem is yet to be addressed -
You must be logged in to vote
