search

google / TensorNetwork

A library for easy and efficient manipulation of tensor networks.
Apache License 2.0
1.8k stars 355 forks source link

Parallelism Contractors #963

Open gmagnifico opened 2 years ago

gmagnifico commented 2 years ago

Hi, I have a code in which the most computationally expensive part is the contraction of the whole network, that is composed by a number of tensors usually larger than 15/20. I am using the functions "contractors.auto" and the jax backend, but I notice that there is no parallelism. The code sees the GPU but its usage is always at 0%. On CPU the code uses just one core.

What are the best ways to speed up the computation?

Thank you very much, Giuseppe

mganahl commented 2 years ago

hi @gmagnifico. Can you post a minimal example code here? Did you set up jax properly to use the GPU?