Aprendizado por reforço para alinhamento de LLMs estudo comparativo de estratégias de recompensa pós-destilação

Abstract

This Course Completion Report aims to bring together the results of my journey to become an expert in Alignment and Optimization of LLMs via Reinforcement Learning. An illustration and its narrative describe the work periods. The Appendices contain the Delivery Acceptance Terms and the results obtained during each work period.