Offline Regularised Reinforcement Learning for Large Language Models Alignment

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Richemond, Pierre Harvey, Tang, Yunhao, Guo, Daniel, Calandriello, Daniele, Azar, Mohammad Gheshlaghi, Rafailov, Rafael, Pires, Bernardo Avila, Tarassov, Eugene, Spangher, Lucas, Ellsworth, Will, Severyn, Aliaksei, Mallinson, Jonathan, Shani, Lior, Shamir, Gil, Joshi, Rishabh, Liu, Tianqi, Munos, Remi, Piot, Bilal
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917678302625792
author Richemond, Pierre Harvey
Tang, Yunhao
Guo, Daniel
Calandriello, Daniele
Azar, Mohammad Gheshlaghi
Rafailov, Rafael
Pires, Bernardo Avila
Tarassov, Eugene
Spangher, Lucas
Ellsworth, Will
Severyn, Aliaksei
Mallinson, Jonathan
Shani, Lior
Shamir, Gil
Joshi, Rishabh
Liu, Tianqi
Munos, Remi
Piot, Bilal
author_facet Richemond, Pierre Harvey
Tang, Yunhao
Guo, Daniel
Calandriello, Daniele
Azar, Mohammad Gheshlaghi
Rafailov, Rafael
Pires, Bernardo Avila
Tarassov, Eugene
Spangher, Lucas
Ellsworth, Will
Severyn, Aliaksei
Mallinson, Jonathan
Shani, Lior
Shamir, Gil
Joshi, Rishabh
Liu, Tianqi
Munos, Remi
Piot, Bilal
contents The dominant framework for alignment of large language models (LLM), whether through reinforcement learning from human feedback or direct preference optimisation, is to learn from preference data. This involves building datasets where each element is a quadruplet composed of a prompt, two independent responses (completions of the prompt) and a human preference between the two independent responses, yielding a preferred and a dis-preferred response. Such data is typically scarce and expensive to collect. On the other hand, \emph{single-trajectory} datasets where each element is a triplet composed of a prompt, a response and a human feedback is naturally more abundant. The canonical element of such datasets is for instance an LLM's response to a user's prompt followed by a user's feedback such as a thumbs-up/down. Consequently, in this work, we propose DRO, or \emph{Direct Reward Optimisation}, as a framework and associated algorithms that do not require pairwise preferences. DRO uses a simple mean-squared objective that can be implemented in various ways. We validate our findings empirically, using T5 encoder-decoder language models, and show DRO's performance over selected baselines such as Kahneman-Tversky Optimization (KTO). Thus, we confirm that DRO is a simple and empirically compelling method for single-trajectory policy optimisation.
format Preprint
id arxiv_https___arxiv_org_abs_2405_19107
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Offline Regularised Reinforcement Learning for Large Language Models Alignment
Richemond, Pierre Harvey
Tang, Yunhao
Guo, Daniel
Calandriello, Daniele
Azar, Mohammad Gheshlaghi
Rafailov, Rafael
Pires, Bernardo Avila
Tarassov, Eugene
Spangher, Lucas
Ellsworth, Will
Severyn, Aliaksei
Mallinson, Jonathan
Shani, Lior
Shamir, Gil
Joshi, Rishabh
Liu, Tianqi
Munos, Remi
Piot, Bilal
Machine Learning
Artificial Intelligence
The dominant framework for alignment of large language models (LLM), whether through reinforcement learning from human feedback or direct preference optimisation, is to learn from preference data. This involves building datasets where each element is a quadruplet composed of a prompt, two independent responses (completions of the prompt) and a human preference between the two independent responses, yielding a preferred and a dis-preferred response. Such data is typically scarce and expensive to collect. On the other hand, \emph{single-trajectory} datasets where each element is a triplet composed of a prompt, a response and a human feedback is naturally more abundant. The canonical element of such datasets is for instance an LLM's response to a user's prompt followed by a user's feedback such as a thumbs-up/down. Consequently, in this work, we propose DRO, or \emph{Direct Reward Optimisation}, as a framework and associated algorithms that do not require pairwise preferences. DRO uses a simple mean-squared objective that can be implemented in various ways. We validate our findings empirically, using T5 encoder-decoder language models, and show DRO's performance over selected baselines such as Kahneman-Tversky Optimization (KTO). Thus, we confirm that DRO is a simple and empirically compelling method for single-trajectory policy optimisation.
title Offline Regularised Reinforcement Learning for Large Language Models Alignment
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2405.19107