LightGBM icon indicating copy to clipboard operation
LightGBM copied to clipboard

Published 20 hours ago • verified publisher icon microsoft

Add quantized training

Open shiyu1994 opened this issue 2 years ago • 5 comments

Summary

Add quantized training algorithm, which accelerates training on both CPU and GPU. See our latest NeurIPS 2022 paper (https://neurips.cc/virtual/2022/poster/53370).

shiyu1994 avatar Nov 25 '22 16:11 shiyu1994

Interesting! Is there a link to the poster materials that doesn't require creating an account?

image

jameslamb avatar Nov 26 '22 00:11 jameslamb

Maybe this link can be accessed without registration to the conference. https://nips.cc/Conferences/2022/Schedule?showEvent=53370

shiyu1994 avatar Nov 27 '22 14:11 shiyu1994

Yep that worked, thank you! Very exciting!

jameslamb avatar Nov 28 '22 03:11 jameslamb

Is there an update regarding the above?

replacementAI avatar Jan 07 '23 04:01 replacementAI

Is there an update regarding the above?

It seems like that it's not updated.

AllenSun1024 avatar Mar 16 '23 10:03 AllenSun1024

Given that quantized training is now supported for CPU and CUDA, I think this issue can be closed.

I'm closing this. @guolinke @shiyu1994 if I'm wrong about that please do re-open it and add a comment explaining what else still needs to be done.

Thanks for the quick translation from your research work into working, released code in this project that its users can get with just a pip install. Pretty exciting!!!

jameslamb avatar Sep 08 '23 02:09 jameslamb

Ah sorry, did not realize #5933 was not merged yet. I'll re-open this.

jameslamb avatar Sep 08 '23 15:09 jameslamb

@jameslamb No worries. I will continue my work in #5933.

shiyu1994 avatar Sep 08 '23 15:09 shiyu1994

I got too excited about closing old issues, sorry 😆

jameslamb avatar Sep 08 '23 15:09 jameslamb