WebGym documentation

• Getting Started

Environment

• Rollout Server Env
  • Prerequisites
  • Create Environment
  • Install Dependencies
  • Verify Installation
  • Launch the Server
• RL Pipeline Env
  • Prerequisites
  • Create Environment
  • Install Dependencies
  • Verify Installation
  • Environment Variables
  • Troubleshooting

Rollout Server

• Deployment
  • Quick Start
  • Deployment Options
  • Architecture
  • Starting Components Individually
  • Verifying the Server
  • API Usage
  • Security
  • Troubleshooting
  • Deployment Files
  • Next Steps
• Multi-node Deployment
  • Architecture Overview
  • Prerequisites
  • Quick Start
  • Deployment Modes
  • Configuration Options
  • Service Discovery Flow
  • Stopping & Restarting
  • Troubleshooting
  • Scaling Examples
  • Next Steps
• Rollout Server Code Details (Optional Read)
  • Architecture Overview
  • Module Structure
  • Components
  • Integration with WebGym
  • Multi-Node Deployment
  • Troubleshooting

Launcher Scripts

• Entry Script: run.sh
  • Quick Start
  • Required Arguments
  • RL Phases
  • Options
  • Argument Combinations
  • Config Overrides
  • Multi-Node
  • Built-in Configuration
  • Helper Functions
  • Environment Variables
  • Stopping the Script
  • Troubleshooting
• Multi-Node Mode: run.sh
  • Setup
  • Examples
  • Coordination Protocol
  • Fault Tolerance
• Configuration Reference
  • WebGym Configs
  • DeepSpeed Configs
• Rollout Script: rollout.py (Optional Read)
  • Overview
  • Entry Point
  • Key Components
  • Metrics
  • Output
• Update Script: update_prepare.py (Optional Read)
  • Overview
  • Entry Point
  • Key Components
  • Data Preparation Pipeline
  • LLaMA-Factory Configuration
  • Configuration
  • Output
  • Next Steps

Analysis Tools

• Viewer
  • Trajectory Viewer
• Visualizer
• Task Monitor

RL Pipeline (Optional Read)

• RL Pipeline Overview
  • Component Summary
  • Data Flow
• Models
  • Module Structure
  • WebAgent
  • Evaluator
  • Configuration
• Context Management
  • Module Structure
  • ContextManager
  • Interaction Modes
  • Parsers
• Replay Buffer
  • Module Structure
  • ReplayBuffer
  • Data Components
• Rollout Collection
  • Module Structure
  • AsyncWebGym
  • Evaluation Integration
• Policy Configuration
  • Module Structure
  • WebAgent
  • Model Factory
  • Base Classes
• WandB Logger
  • Module Structure
  • Key Functions
  • Configuration