Distributed reinforcement learning on a single machine with multiple GPUs

LiuShuai26 / Distributed-RL

Distributed DRL by Ray and TensorFlow Tutorial.

9 stars 0 forks source link

Distributed reinforcement learning on a single machine with multiple GPUs #1

Open namjiwon1023 opened 3 years ago

namjiwon1023 commented 3 years ago

@LiuShuai26 Thank you for your contribution! Very helpful to me. May I ask the difference between this code and APE-X and whether this code can be used in a single machine multi-GPUs environment. Waiting for your reply, thank you again for your contribution!

LiuShuai26 commented 3 years ago

This code is a general framework simply implement by ray which is similar to APEX. It doesn't have prioritized experience replay implementation to compare to APEX. It doesn't support single machine multi-GPUs environment. In multi-GPUs set up, gradient allreduce tool like horovod is needed.