使用finetune_hf.py进行65条数据的微调,报cuda out of memory,是需要添加什么配置的吗? #1085
Closed
zhaohai7206
started this conversation in
General
Replies: 2 comments 2 replies
-
看晚上有人使用过python finetune_hf.py AdvertiseGen chatglm/chatglm3-6b configs/lora.yaml output,后面参数多了一个output,但是我执行的会报下面的问题,并且output目录没有任何生成任何文件 |
Beta Was this translation helpful? Give feedback.
1 reply
-
16G可能不够的话,你可以降低一下max token 16G太极限了 |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
我这是两张4080卡,都是16G的,执行环境也是在conda,按照requirement里面进行的python包安装,chatglm3-6模型,从https://hf-mirror.com/THUDM/chatglm3-6b/tree/main上面下载的,但是执行python finetune_hf.py ../../AdvertiseGen chatglm/chatglm3-6b configs/lora.yaml 进行微调的时候就会报oom

请问下,是需要配置什么的吗,代码文件没有任何修改的
Beta Was this translation helpful? Give feedback.
All reactions