When I train a network using pytorch, sometimes I do not what a tensor participate in gradient backward. But however, I cannot find a function that is similar to “detach()” in julia.Flux. So when I use julia to train a network, how can I make a tensor do not participate in gradient backward? …

Which function in julia.Flux is similar to "detach()" in pytorch?

CarloLucibello August 29, 2024, 8:27am 2

Probably ignore_derivatives is what you are looking for.

3 Likes