Deciphering Trajectory-Aided LLM Reasoning: An Optimization Perspective

Junnan Liu, Hongwei Liu, Linchen Xiao +5

We propose a novel framework for comprehending the reasoning capabilities of large language models (LLMs) through the perspective of meta-learning. By conceptualizing reasoning tra…