Skip to main content
This section covers the experimental features built around the UnifiedTrainer and its ecosystem.

​
Tutorials

Solver-judge workflow

Build a multi-agent solver-judge workflow and train it end-to-end

​
Unified workflow trainer

Unified trainer

End-to-end training loop architecture and lifecycle

Backend protocol

Backend interface contract used by the unified trainer

Configuration

Config structure and forwarding behavior

​
Algorithms

Advantage estimator

rLLM-native advantage estimation and customization

Pre-computing advantage

Step-level precomputed training signals and mixed-mode training

​
Monitoring

rLLM UI

Real-time monitoring for agent training and evaluation