gflowpo-generative-flow-network - Optimize LLM prompts with GFlowPO
Optimizes LLM prompts through iterative candidate generation, example-based scoring, diversity-preserving replay, dynamic memory updates, and held-out validation.
Tags
Updated: 2026-09-28Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Generate diverse prompt candidates
- Score candidates on examples
- Update replay buffer
- Update priority queue
- Refresh meta-prompt references
- Validate on held-out examples
- Report optimization trace
Inputs
- Target task
- Input/output examples
- Held-out test examples
- Initial or current prompt
- Evaluation metric
Outputs
- Optimized prompt
- Runner-up prompts
- Candidate scores
- Test accuracy
- Optimization trace
Requirements
- LLM prompt generation support
- LLM task execution support