FutureWarning: `torch.cpu.amp.autocast(args...)` is deprecated.

[fumize]

sorry,the newest version show this Warning ,is this can ignore?
just feel Loss is unstable now

[uYouUs]

Yes, you can ignore that warning. Its a warning to developers, letting them know that the code has changed. To ask them to change how they are using it to the new way as it will not be supported the way it is in the future, producing a real error then.
For now, so long as it still works, it works.

As for the loss, that's with your training settings and dataset. I cant help on that without more information. What do you mean by unstable, what settings are you using, and what is your dataset like?