Nan in Loss/Score

[HanshuYAN]

Dear Authors,

Thanks for the open-source codes. I run the cifar10 experiment on a single-card with the following code:

• python main.py -cc configs/default_cifar10.txt -sc configs/specific_cifar10.txt --root $ROOT --mode train --workdir work_dir/cifar10 --n_gpus_per_node 1 --training_batch_size 64 --testing_batch_size 64 --sampling_batch_size 64

However, the loss is always nan (error in the line 61 of losses.py). I check the output values. The score values in line 45 are all NAN. Do you know how to solve this issue?

regards,
Hanshu

[Superviro]

Same problem here, any solutions?