BobaZooba/xllm

DeepSpeed on a sigle GPU

Closed this issue · 0 comments

You need to add the ability to use DeepSpeed on a single GPU. Due to offloading, there will be a reduction in memory consumption during training