fix save checkpoint and best model process

shin04 / contrastive-learning-for-audio

2 stars 0 forks source link

Closed shin04 closed 3 years ago

shin04 commented 3 years ago

saving both weights at /model/...
saving two type best model (per epoch and per 100 step)
The loss at the checkpoint is strange.
- Now: save sum of losses at an epoch (loss_epoch)
- Update: save average of losses at an epoch (loss_epoch / len(dataloader))