Base Reward Manager for GenEnv training. >>> USER CUSTOMIZATION REQUIRED <<< Replace the `compute_reward` method with your domain-sp

Reward Manager for Action-based tasks (e.g., web navigation). This implementation checks if the model's predicted actions match the

Environment Generator Worker using vLLM for efficient inference. This worker is responsible for generating new training tasks based on

Reward Manager for Tool/API Calling tasks. This implementation checks if the model's predicted tool calls match the expected ground