DeepSpeedExamples icon indicating copy to clipboard operation
DeepSpeedExamples copied to clipboard

how to use zero-offload?

Open xdnjust opened this issue 1 year ago • 2 comments

Hi, I am trying to train a GPT-2 model using "DeepSpeed-Chat” code. Bur in step 1, when I use the "--offload", I got a error. below is the problem:

image

xdnjust avatar Apr 21 '23 01:04 xdnjust

not supported now

Modas-Li avatar Apr 21 '23 11:04 Modas-Li

Zero offload isn’t supported yet, we’re actively working on this though. Will update when it’s released :)

Feel free to also follow our Twitter account for updates like this: https://twitter.com/msftdeepspeed

jeffra avatar Apr 22 '23 03:04 jeffra

Hey @jeffra 🙂, is this not implemented only for the hybrid engine, i.e. would it work as expected if I disable Hybrid Engine and remove the assertion from the main.py scripts?

EikeKohl avatar Apr 28 '23 13:04 EikeKohl

@jeffra Can i use Zero Zero offload?

zhangyanbo2007 avatar May 08 '23 15:05 zhangyanbo2007

@jeffra Why can't we use zero-offload? Can you tell me the specific reason? I can successfully enable zero-offload。

zhihao-chen avatar May 11 '23 07:05 zhihao-chen