VAGEN¶
VAGEN is a multi-turn reinforcement learning framework for training Vision-Language Model (VLM) agents.
Version Switch
You are viewing the VAGEN-Lite documentation. For the previous full version, see VAGEN-Legacy Documentation.
VAGEN-Lite¶
A lightweight reimplementation built on VERL agent-loop for easy customization, fast training, and stable performance.
Documentation¶
- Quick Start - Installation and basic usage
- Configuration - Training configuration reference
- Custom Environment - Build your own environment
- Custom Filter - Preprocess training data
- Custom Metric - Add W&B logging metrics