How can I inference with fp16

[zelabean]

Hello. I would like to infer a model trained with a trainer (precision=16) with fp16.
What should I do?

[rohitgr7]

trainer = Trainer(precision=16)
trainer.test(...)

this doesn't work?

[zelabean]

I want use it on deployment environment. so just i want just using forward function like

with torch.no_grad():
model.forward(input)

don't wanna use Trainer().

thank you for answer :)