RL with Multiple Agents in distinct environments on AWS

0

Hello, I have a setup using Ray to setup RL agents and their independent environments on separate instances (20 agents sharing 6 GPUs) using Pytorch on a private cluster. Any thoughts on how to go about setting this up on Amazon AWS? I was able to get one agent running my custom RL code and custom environment on Sagemaker. It is not necessarily distributed RL because each agent runs its own environment and learns individually (only using a shared folder). Any suggestions would be greatly appreciated as I am in a time crunch and looking for additional places to run my simulations. Thank you!!!

lkerb
asked 2 months ago111 views
1 Answer
0

Hello,

From the description, I understand that you are exploring options to run distributed RL on AWS.

You can definitely continue to use SageMaker to achieve your use-case because SageMaker RL supports multi-core and multi-instance distributed training. Depending on your use case, training and/or environment rollout can be distributed. Please refer to the doc to learn more about the same.

Further, you can also refer to the below mentioned blogs that aim to illustrate a similar use-case in SageMaker which should be a good starting point to test the overall setup.

Blog1 Deploying reinforcement learning in production using Ray and Amazon SageMaker

Blog2 Scaling your AI-powered Battlesnake with distributed reinforcement learning in Amazon SageMaker

Therefore, I would recommend exploring the above references to analyze the approach to deploy distributed RL on SageMaker and then make necessary changes to adapt the same for your use-case.

AWS
Harsha
answered 2 months ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.

Guidelines for Answering Questions