Self-Play with Adversarial Critic: Provable and Scalable Offline Alignment for Language Models

• URL: http://arxiv.org/abs/2406.04274v1
• Date: Thu, 6 Jun 2024 17:23:49 GMT
• Title: Self-Play with Adversarial Critic: Provable and Scalable Offline Alignment for Language Models
• Authors: Xiang Ji, Sanjeev Kulkarni, Mengdi Wang, Tengyang Xie,
• Abstract summary: This work studies the challenge of aligning large language models (LLMs) with offline preference data. We propose SPAC, a new offline preference optimization method with self-play, inspired by the on-average pessimism technique from the offline RL literature.
• Score: 44.38073745307387
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: This work studies the challenge of aligning large language models (LLMs) with offline preference data. We focus on alignment by Reinforcement Learning from Human Feedback (RLHF) in particular. While popular preference optimization methods exhibit good empirical performance in practice, they are not theoretically guaranteed to converge to the optimal policy and can provably fail when the data coverage is sparse by classical offline reinforcement learning (RL) results. On the other hand, a recent line of work has focused on theoretically motivated preference optimization methods with provable guarantees, but these are not computationally efficient for large-scale applications like LLM alignment. To bridge this gap, we propose SPAC, a new offline preference optimization method with self-play, inspired by the on-average pessimism technique from the offline RL literature, to be the first provable and scalable approach to LLM alignment. We both provide theoretical analysis for its convergence under single-policy concentrability for the general function approximation setting and demonstrate its competitive empirical performance for LLM alignment on a 7B Mistral model with Open LLM Leaderboard evaluations.

Related papers

• Preference Elicitation for Offline Reinforcement Learning [59.136381500967744]
  We propose Sim-OPRL, an offline preference-based reinforcement learning algorithm. Our algorithm employs a pessimistic approach for out-of-distribution data, and an optimistic approach for acquiring informative preferences about the optimal policy.
  arXiv  Detail & Related papers  (2024-06-26T15:59:13Z)
• SAIL: Self-Improving Efficient Online Alignment of Large Language Models [56.59644677997827]
  Reinforcement Learning from Human Feedback is a key method for aligning large language models with human preferences. Recent literature has focused on designing online RLHF methods but still lacks a unified conceptual formulation. Our approach significantly improves alignment performance on open-sourced datasets with minimal computational overhead.
  arXiv  Detail & Related papers  (2024-06-21T18:05:35Z)
• The Importance of Online Data: Understanding Preference Fine-tuning via Coverage [25.782644676250115]
  We study the similarities and differences between online and offline techniques for preference fine-tuning. We prove that a global coverage condition is both necessary and sufficient for offline contrastive methods to converge to the optimal policy. We derive a hybrid preference optimization algorithm that uses offline data for contrastive-based preference optimization and online data for KL regularization.
  arXiv  Detail & Related papers  (2024-06-03T15:51:04Z)
• Self-Augmented Preference Optimization: Off-Policy Paradigms for Language Model Alignment [104.18002641195442]
  We introduce Self-Augmented Preference Optimization (SAPO), an effective and scalable training paradigm that does not require existing paired data. Building on the self-play concept, which autonomously generates negative responses, we further incorporate an off-policy learning pipeline to enhance data exploration and exploitation.
  arXiv  Detail & Related papers  (2024-05-31T14:21:04Z)
• Value-Incentivized Preference Optimization: A Unified Approach to Online and Offline RLHF [80.32171988565999]
  We introduce a unified approach to online and offline RLHF -- value-incentivized preference optimization (VPO) VPO regularizes the maximum-likelihood estimate of the reward function with the corresponding value function. Experiments on text summarization and dialog verify the practicality and effectiveness of VPO.
  arXiv  Detail & Related papers  (2024-05-29T17:51:42Z)
• Learning Goal-Conditioned Policies from Sub-Optimal Offline Data via Metric Learning [22.174803826742963]
  We address the problem of learning optimal behavior from sub-optimal datasets for goal-conditioned offline reinforcement learning. We propose the use of metric learning to approximate the optimal value function for goal-conditioned offline RL problems. We show that our method estimates optimal behaviors from severely sub-optimal offline datasets without suffering from out-of-distribution estimation errors.
  arXiv  Detail & Related papers  (2024-02-16T16:46:53Z)
• Linear Alignment: A Closed-form Solution for Aligning Human Preferences without Tuning and Feedback [70.32795295142648]
  Linear alignment is a novel algorithm that aligns language models with human preferences in one single inference step. Experiments on both general and personalized preference datasets demonstrate that linear alignment significantly enhances the performance and efficiency of LLM alignment.
  arXiv  Detail & Related papers  (2024-01-21T10:46:23Z)
• Revisiting the Linear-Programming Framework for Offline RL with General Function Approximation [24.577243536475233]
  offline reinforcement learning (RL) concerns pursuing an optimal policy for sequential decision-making from a pre-collected dataset. Recent theoretical progress has focused on developing sample-efficient offline RL algorithms with various relaxed assumptions on data coverage and function approximators. We revisit the linear-programming framework for offline RL, and advance the existing results in several aspects.
  arXiv  Detail & Related papers  (2022-12-28T15:28:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.