mengzi-retrieval-lm icon indicating copy to clipboard operation
mengzi-retrieval-lm copied to clipboard

About the compute resources

Open WorldHellooo opened this issue 2 years ago • 2 comments

Thanks for making your work public! Want to know how many computing resources were used for training and retrieval when you train the GPT-125M model?

WorldHellooo avatar Feb 06 '23 13:02 WorldHellooo

I am curious as well.

daniellefisla avatar Feb 22 '23 02:02 daniellefisla

We used two 8*A100 40G servers for training, one as an index server and one as a training server. If you want to train a larger model, 1 index server can also be used, just increase the number of training servers

Ag2S1 avatar Mar 20 '23 14:03 Ag2S1