Question about training with mixed precision.

[eightmusic]

Hello, I'm using tutorial_train. py, and will be "trainer = pl. Trainer (gpus = 1, precision = 32, callbacks = [logger])" in the precision is set to 16, for mixing precision training, However, I found that whether I set it to 16 or 32, the cuda memory and training time will not change, what is the reason?

[kornellewy]

I think model is not optimized for that, stay on 32.