Closed Isjacky closed 5 months ago
python main.py --dataset CIFAR100 --data_split_file ./data_split/CIFAR100_split_cn10_tn4_cet20_s2571.pkl --num_glob_iters 40 --local_epochs 400 --lr 1e-3 --flow_lr 5e-3 --k_loss_flow 0.5 --k_flow_lastflow 0.1 --flow_explore_theta 0.1 --fedprox_k 0.001 跑出来的结果出现很多nan pickle.csv
看起来是normalizing flow的训练loss出现nan,NF的训练比较tricky,不同机器的训练结果可能不同,尝试一下别的参数?
好的,我多尝试一下,谢谢!
python main.py --dataset CIFAR100 --data_split_file ./data_split/CIFAR100_split_cn10_tn4_cet20_s2571.pkl --num_glob_iters 40 --local_epochs 400 --lr 1e-3 --flow_lr 5e-3 --k_loss_flow 0.5 --k_flow_lastflow 0.1 --flow_explore_theta 0.1 --fedprox_k 0.001 跑出来的结果出现很多nan pickle.csv