Checkpoints are overwritten automatically

callbacks
1

Hi,

I’m new to pytorch lightning and trying to use the callback to save model checkpoints. But it only saved one file and overwrote it automatically. The code is the following:

checkpoint_callback = ModelCheckpoint(
    dirpath='./logs/', every_n_epochs=5, save_on_train_epoch_end=True, filename=None)
model = Pix2Pix(len(forcings), 1)

trainer = pl.Trainer(max_epochs=100, gpus=1,
                     gradient_clip_val=1., callbacks=[checkpoint_callback])
trainer.fit(model, loader)

If my understanding correctly, it should save checkpoints every 5 epochs so the files should be epoch=4.ckpt, epoch=9.ckpt, epoch=14.ckpt, etc., But I only get one file and it seems the previous ones are all overwritten.

Also the file name has apostrophes with it. Here is a screenshot of my current log directory.

image
image878×96 6.94 KB

Any suggestions would be appreciated!

2

It seems I missed the top_k parameter.