To test the model

[wxthu]

Hi, when I finished the task of training model using customized datasets and tested with command "python generate.py checkpoint/010000.model", I encountered the error as following:

""""
generator.load_state_dict(torch.load(args.path)['g_running'])

KeyError: "g_running"
""""
I cannot understand. Could you please give me a help ? Thx

[rosinality]

You need to use just generator.load_state_dict(torch.load(args.path)). I don't remember why I have used different checkpoint structures.

[wxthu]

Thanks for your reply and when I run "python train.py --mixing --loss r1 --sched LMDB_PATH" using my customized datasets, no checkpoints generated and I am confused. (ps: I am newcomer for this field)

[wxthu]

Also, I wanna know how to accelerate the training process in the command line ? It seems so slow....

[rosinality]

Checkpoints should have been saved at every 10,000 iterations or phase (resolution) changes.

And as this implementation uses nn.DataParallel for multi gpu training so it is not very efficient. You may need to consider the implementations that uses nn.parallel.DistributedDataParallel.