Task-Graph GAR Planner
What it is
A framework for training a planner that decomposes complex mathematical problems into Directed Acyclic Graphs (DAGs) for parallel execution. It uses a two-stage approach: Supervised Fine-Tuning (SFT) for JSON structure learning followed by Group Relative Policy Optimization (GRPO) with multi-objective rewards.
The problem it solves
Standard LLM reasoning often struggles with complex decomposition and lacks a mechanism to optimize for parallel execution paths.
What you could use it for
- Decomposing complex math word problems into parallelizable sub-tasks
- Automated task planning for multi-agent systems
- Generating structured execution graphs from natural language prompts
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.