Getting Started

To get started with WebGym, you need to prepare two types of machines:

• CPU machines for hosting rollout servers (OmniBoxes)

• GPU machines for hosting the RL training pipeline

If your GPU machine contains >128 CPUs and >256GB CPU RAM, it is also possible to host CPU services on these GPU machines locally.

First, set up the environments on both machine types. Refer to Rollout Server Env for rollout server environment setup and RL Pipeline Env for RL pipeline environment setup.

Once the environments are ready, host the rollout server and make sure it is properly tested. Refer to Deployment for a quickstart guide and Rollout Server Code Details (Optional Read) for the full architecture and API details.

Next, launch the training script on the GPU machine. Refer to Entry Script: run.sh for the main entry script and Configuration Reference for configuration options.

After RL training completes, analyze the collected trajectories to evaluate performance. Refer to the Analysis Tools section for the available analysis and visualization tools.

For developers looking to understand or extend the system internals, refer to RL Pipeline Overview for details on RL pipeline components and development.