Facico / Chinese-Vicuna

Chinese-Vicuna: A Chinese Instruction-following LLaMA-based Model —— 一个中文低资源的llama+lora方案,结构参考alpaca
https://github.com/Facico/Chinese-Vicuna
Apache License 2.0
4.14k stars 425 forks source link

finetune_deepspeed启动运行[ERROR] [launch.py:324:sigkill_handler] #218

Open grantchenhuarong opened 1 year ago

grantchenhuarong commented 1 year ago

image

环境是4090,python3.10.11,所有依赖按照requirements.txt安装, cuda11.7, nvidia 12.1

请教一下如何才能正常启动deepspeed呀?

grantchenhuarong commented 1 year ago

ds_report coredump image

grantchenhuarong commented 1 year ago

dmesg信息查看 https://github.com/microsoft/DeepSpeed/issues/2632 image

grantchenhuarong commented 1 year ago

唉, 4090容不下,内存也装载不了,OS这时候干预了。。。

image

Facico commented 1 year ago

你这个看起来是爆内存了,你跑的多大的模型?不过对于训练lora来说deepspeed省的显存并不明显 4090跑这个脚本13B应该是可以放下的,不过batch size得设置小一点