torch.no_grad()

Memory efficient (without back propagation)

validation inner loop를 이 context manager로 감싸서 gradient calculation을 방지

자동으로 gradient를 트래킹하지 않는다

The primary difference between torch.no_grad() and torch.inference_mode() in PyTorch is that torch.no_grad() temporarily disables gradient calculation, while torch.inference_mode() not only disables gradient computation but also optimizes specific operations for inference.

no_grad — PyTorch 2.1 documentation

Community

https://pytorch.org/docs/stable/generated/torch.no_grad.html

`torch.no_grad()`

Memory efficient (without back propagation)

Recommendations