search

osudrl / apex

A continuous deep reinforcement learning framework for robotics
MIT License
60 stars 31 forks source link

Fixes a ray/pytorch bottleneck #13

Closed siekmanj closed 5 years ago

siekmanj commented 5 years ago

Stops pytorch from using more than one core when sampling in parallel. This seems to help performance a lot, especially on vLab (2x-4x speedup).

siekmanj commented 5 years ago

(Yesh, I think this would probably help your training setup considerably)