ChatGLM-6B icon indicating copy to clipboard operation
ChatGLM-6B copied to clipboard

Published 20 hours ago • verified publisher icon THUDM

[Help] 请问如何实现多卡微调？

Open pyy1988 opened this issue 2 years ago • 2 comments

Is there an existing issue for this?

  • [X] I have searched the existing issues

Current Behavior

/

Expected Behavior

No response

Steps To Reproduce

/

Environment

/

Anything else?

No response

pyy1988 avatar Apr 28 '23 14:04 pyy1988

See https://github.com/THUDM/ChatGLM-6B/tree/main/ptuning#finetune （？

Btlmd avatar Apr 29 '23 07:04 Btlmd

See https://github.com/THUDM/ChatGLM-6B/tree/main/ptuning#finetune （？

官方给的多卡加载模型方法load_model_on_gpus一直报错，sh配置文件中CUDA_VISIBLE_DEVICES=0添加为0,1，两卡都用上了，ptuning v2微调int8模型，其他参数不变的情况下，每张卡卡占用接近24g，看起来比官方给出的最低9G要高出不少

pyy1988 avatar Apr 30 '23 00:04 pyy1988

找到原因

pyy1988 avatar May 09 '23 01:05 pyy1988

找到原因

你好，你这边是怎么解决的啊？我多卡ptuning也遇到类似问题，两张卡的情况下平均每张卡占用的显存比我单卡ptuning时高不少

tolecy avatar May 16 '23 03:05 tolecy

@tolecy 请问这边您是如何解决的，我的情况和您类似，多卡反而比单卡高不少

niuhuluzhihao avatar Jun 16 '23 15:06 niuhuluzhihao