Skip to content

GitLab

Open
Created by xdchenhao@xdchenhao

stage3_gather_16bit_weights_on_model_save为true保存模型会爆内存，有解决方案么

训练qwen-7b-chat时deepspeed.json里面stage3_gather_16bit_weights_on_model_save设置为true会爆内存 image