Torch acceleration

After profiling we found several bottlenecks in the code and fixed them.

Merge request reports

Loading