How to continue to train with the checkpoint?

diaoenmao / HeteroFL-Computation-and-Communication-Efficient-Federated-Learning-for-Heterogeneous-Clients

[ICLR 2021] HeteroFL: Computation and Communication Efficient Federated Learning for Heterogeneous Clients

MIT License

151 stars 33 forks source link

How to continue to train with the checkpoint? #13

Closed HideLakitu closed 3 hours ago

HideLakitu commented 3 days ago

Hi, Just try to reproduct experment now.

I Open config.yml, set resume mode in it likeresume_mode: '0_CIFAR10_label_resnet18_1_10_0.1_non-iid-2_dynamic_a1-b1-c1_gn_0_0_checkpoint.pt' (the default value is 0), which the checkpoint file is automatically generated in the folder output/model last time, then execute the model again?

But this seems didn't work, what should I do

diaoenmao commented 3 hours ago

I am not sure resuming checkpoint is working in this code.