[💡SUG] Will multi-GPU be supported in the next version?

[sxhysj]

Will multi-GPU be supported in the next version?

[Sherry-XLL]

@sxhysj Sorry, considering the complexity of implementation, RecBole's next version still does not support multi-GPU training. We have taken multi-GPU training into consideration. If you have some ideas or relevant papers, comments and suggestions are very welcome as always.

[dxjjhm]

I suggest you to learn the method of DJL (DeepJavaLibrary https://djl.ai/ ), it could use all the gpus of single machine.

[sxhysj]

Any updates on this?

[2017pxy]

@sxhysj Sorry, the implementation of this feature is a bit complex in our framework and it still takes some time to add it.

BTW, @juyongjiang has given his implementation in #961, which works well for sequential recommendation models. You can refer it and try to modify our code by yourself.

[Sherry-XLL]

Hello @sxhysj, we have released the 1.1.1 version of RecBole on Oct 5 2022, and the distributed training and evaluation is supported in our framework now. You can refer to our documentation for more details and running examples.

Thank you very much for your attention and kindly advice! The suggestions and feedback from open source users are our driving force for continuous updating. We welcome you very much to use our library again and give your valuable comments.