wellinxu / LLM_Custome

在中文开源大模型的基础上进行定制化的微调,拥有自己专属的语言模型。
Apache License 2.0
41 stars 7 forks source link

放出的lora权重版本是否和基座模型一致 #3

Open kangbrilliant opened 9 months ago

kangbrilliant commented 9 months ago

您好,按照requirements.txt安装好依赖环境后,从huggingface.co下载好「chatglm-6b」和「lora」权重,运行『chat_server.py』后显示 hidden-size mismatch错误,([8192, 8, 1]) vs ([12288, 8]).

wellinxu commented 1 month ago

抱歉,好久没上,是基于初版的chatglm-6b训练的