OptimalScale/LMFlow

lora training - Input:Instruction

Yyyyyyxh opened this issue · 1 comments

您好,利用工程自带数据集对Llama7b进行Lora训练,数据集如下图所示
截屏2023-10-19 10 40 09
训练后app.py参数如下设置
截屏2023-10-19 10 39 53
对话显示如下所示

截屏2023-10-19 10 39 19

Thanks for your interest in LMFlow! server/app.py didn't take prompt format into consideration. You may first use examples/chatbot.py with --prompt_structure "Input: Instructions: {input_text} Output:". This will alleviate the problem. To make server/app.py work, we can do the same string formatting by modifying the prompt format in line 116-123 (https://github.com/OptimalScale/LMFlow/blob/main/service/app.py#L116).

Hope that can solve your issue 🙏