with_no_grad: Temporarily modify gradient recording.

In torch: Tensors and Neural Networks with 'GPU' Acceleration

Temporarily modify gradient recording.

Description

Temporarily modify gradient recording.

Usage

with_no_grad(code)

local_no_grad(.env = parent.frame())

Arguments

code	code to be executed with no gradient recording.
.env	The environment to use for scoping.

Functions

• local_no_grad(): Disable autograd until it goes out of scope

Examples

if (torch_is_installed()) {
x <- torch_tensor(runif(5), requires_grad = TRUE)
with_no_grad({
  x$sub_(torch_tensor(as.numeric(1:5)))
})
x
x$grad
}

torch documentation built on May 29, 2024, 9:54 a.m.