Reinforcement Learning Improves LLM Accuracy and Reasoning in Disease Classification from Radiology Reports

Create a generic template for using GRPO (Group Relative Policy Optimization) to align small LLMs for specialized classification tasks. This will demonstrate how to boost accuracy in domain-specific tasks without sacrificing base logic.

Suggested repo: grpo-align

"Accuracy-focused RL alignment for small models."

Estimated effort: 110h