All geek questions in one place

Implement memory memory method for pytorch

I want to implement the storage method described in this article Learning deep networks with sublinear memory cost . Code that implements this method for the mxnet framework can be found here .

My question is that this way of storing memory is suitable for symbolic graphs like teano and Tensorflow, but the graphs in pytorch are dynamic, so can this method be applied for pytorch?

+4
source share

No one has answered this question yet.

See related questions:

Source: https://habr.com/ru/post/1686356/

More articles:


All Articles