CLGo
CLGo copied to clipboard
Multiple GPU training
@liuruijin17 Thank you very much for the open source, I currently hope to train more GPU, found that the fetchsize and chunk size seem to need to be adjusted synchronously, I don't know how this needs to be changed
Sorry for the very late response, my GPU resource is limited so this version is only tested on a single-GPU machine. But, I would update a multi-GPU branch to fix it. It may need time to train and test, please be patient! Thanks for your interest again!
@liuruijin17 Thank you very much for your reply. Currently, there are some problems with the multi-GPU I tested, and there are some differences with the distributed GPU I tried before. Could you spare some time to help me check, or the training efficiency will be too low