Runfa Chen comments

Repositories
Issues
Comments

Results 3 comments of


                                            Runfa Chen

设置了--light为True依旧拯救不了cuda memory不够用，有什么挽救办法吗？

可以尝试调小--n_res和--n_dis

Training collapses after 72k iterations

It may be that there is a phenomenon of model collapse. The training schedule can be adjusted according to the dataset, and the checkpoint can be saved for manual inspection...

why normalize_weight need div torch.mean(x)?

ce = nn.CrossEntropyLoss(reduction='none')(predict_prob_source, label_source) And why feed after_sotmax(predict_prob_source) to nn.CrossEntropyLoss? This criterion has combined LogSoftmax and NLLLoss.