A Note on Hybrid Online Reinforcement and Imitation Learning for LLMs: Formulations and Algorithms

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Li, Yingru, Li, Ziniu, Liu, Jiacai
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908736879067136
author Li, Yingru
Li, Ziniu
Liu, Jiacai
author_facet Li, Yingru
Li, Ziniu
Liu, Jiacai
contents We present a unified framework for Large Language Model (LLM) fine-tuning that integrates Imitation Learning and Reinforcement Learning. By analyzing the gradient of a composite objective combining trajectory-level KL divergence with task rewards, we derive a natural decomposition into two components: (1) an analytically computable Dense Gradient for token-level imitation, and (2) a Monte Carlo estimated Sparse Gradient for long-horizon reward optimization. The Dense Gradient admits a closed-form logit-level formula, enabling efficient GPU implementation.
format Preprint
id arxiv_https___arxiv_org_abs_2512_23097
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Note on Hybrid Online Reinforcement and Imitation Learning for LLMs: Formulations and Algorithms
Li, Yingru
Li, Ziniu
Liu, Jiacai
Machine Learning
Artificial Intelligence
Computation and Language
We present a unified framework for Large Language Model (LLM) fine-tuning that integrates Imitation Learning and Reinforcement Learning. By analyzing the gradient of a composite objective combining trajectory-level KL divergence with task rewards, we derive a natural decomposition into two components: (1) an analytically computable Dense Gradient for token-level imitation, and (2) a Monte Carlo estimated Sparse Gradient for long-horizon reward optimization. The Dense Gradient admits a closed-form logit-level formula, enabling efficient GPU implementation.
title A Note on Hybrid Online Reinforcement and Imitation Learning for LLMs: Formulations and Algorithms
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2512.23097