Iterative Nash Policy Optimization: Aligning LLMs with General Preferences via No-Regret Learning
- URL: http://arxiv.org/abs/2407.00617v2
- Date: Sun, 7 Jul 2024 09:51:26 GMT
- Title: Iterative Nash Policy Optimization: Aligning LLMs with General Preferences via No-Regret Learning
- Authors: Yuheng Zhang, Dian Yu, Baolin Peng, Linfeng Song, Ye Tian, Mingyue Huo, Nan Jiang, Haitao Mi, Dong Yu,
- Abstract summary: We propose a novel algorithm, iterative Nash policy optimization (INPO)
Unlike previous methods, INPO bypasses the need for estimating the expected win rate for individual responses.
With an LLaMA-3-8B-based SFT model, INPO achieves a 41.5% length-controlled win rate on AlpacaEval 2.0 and a 38.3% win rate on Arena-Hard.
- Score: 55.65738319966385
- License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
- Abstract: Reinforcement Learning with Human Feedback (RLHF) has achieved great success in aligning large language models (LLMs) with human preferences. Prevalent RLHF approaches are reward-based, following the Bradley-Terry (BT) model assumption, which may not fully capture the complexity of human preferences. In this paper, we explore RLHF under a general preference framework and approach it from a game-theoretic perspective. Specifically, we formulate the problem as a two-player game and propose a novel algorithm, iterative Nash policy optimization (INPO). The key idea is to let the policy play against itself via no-regret learning, thereby approximating the Nash policy. Unlike previous methods, INPO bypasses the need for estimating the expected win rate for individual responses, which typically incurs high computational or annotation costs. Instead, we introduce a new loss objective that is directly minimized over a preference dataset. We provide theoretical analysis for our approach and demonstrate its effectiveness through experiments on various representative benchmarks. With an LLaMA-3-8B-based SFT model, INPO achieves a 41.5% length-controlled win rate on AlpacaEval 2.0 and a 38.3% win rate on Arena-Hard, showing substantial improvement over the state-of-the-art iterative algorithm [Dong et al., 2024] under the BT model assumption. Additionally, our ablation study highlights the benefits of incorporating KL regularization for response length control.
Related papers
- On the Algorithmic Bias of Aligning Large Language Models with RLHF: Preference Collapse and Matching Regularization [33.331389392270665]
preference matching (PM) RLHF is a novel approach that aligns large language models with the preference distribution of the reward model under the Bradley--Terry--Luce/Plackett--Luce model.
Central to our approach is a PM regularizer that takes the form of the negative logarithm of the LLM's policy probability distribution over responses.
For practical implementation, we introduce a conditional variant of PM RLHF that is tailored to natural language generation.
arXiv Detail & Related papers (2024-05-26T07:00:05Z) - Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer [52.09480867526656]
We identify the source of misalignment as a form of distributional shift and uncertainty in learning human preferences.
To mitigate overoptimization, we first propose a theoretical algorithm that chooses the best policy for an adversarially chosen reward model.
Using the equivalence between reward models and the corresponding optimal policy, the algorithm features a simple objective that combines a preference optimization loss and a supervised learning loss.
arXiv Detail & Related papers (2024-05-26T05:38:50Z) - Self-Play Preference Optimization for Language Model Alignment [75.83359213697854]
Recent advancements suggest that directly working with preference probabilities can yield a more accurate reflection of human preferences.
We propose a self-play-based method for language model alignment, which treats the problem as a constant-sum two-player game.
Our approach, dubbed Self-Play Preference Optimization (SPPO), approximates the Nash equilibrium through iterative policy updates.
arXiv Detail & Related papers (2024-05-01T17:59:20Z) - Direct Nash Optimization: Teaching Language Models to Self-Improve with General Preferences [21.5605000515622]
This paper studies post-training large language models (LLMs) using preference feedback from an oracle to help a model iteratively improve over itself.
We introduce Direct Nash Optimization (DNO), a provable and efficient algorithm that marries the simplicity and stability of contrastive learning with theoretical generality from optimizing general preferences.
In our experiments, a resulting 7B parameter Orca-2.5 model achieves the state-of-the-art win-rate against GPT-4-Turbo of 33% on AlpacaE 2.0 (even after controlling for response length), an absolute gain of 26% (7% to 33%) over the initializing model
arXiv Detail & Related papers (2024-04-04T17:56:41Z) - Fine-Tuning Language Models with Reward Learning on Policy [68.70065254564642]
Reinforcement learning from human feedback (RLHF) has emerged as an effective approach to aligning large language models (LLMs) to human preferences.
Despite its popularity, (fixed) reward models may suffer from inaccurate off-distribution.
We propose reward learning on policy (RLP), an unsupervised framework that refines a reward model using policy samples to keep it on-distribution.
arXiv Detail & Related papers (2024-03-28T10:02:10Z) - Nash Learning from Human Feedback [86.09617990412941]
We introduce an alternative pipeline for the fine-tuning of large language models using pairwise human feedback.
We term this approach Nash learning from human feedback (NLHF)
We present a novel algorithmic solution, Nash-MD, founded on the principles of mirror descent.
arXiv Detail & Related papers (2023-12-01T19:26:23Z) - Human-in-the-loop: Provably Efficient Preference-based Reinforcement
Learning with General Function Approximation [107.54516740713969]
We study human-in-the-loop reinforcement learning (RL) with trajectory preferences.
Instead of receiving a numeric reward at each step, the agent only receives preferences over trajectory pairs from a human overseer.
We propose the first optimistic model-based algorithm for PbRL with general function approximation.
arXiv Detail & Related papers (2022-05-23T09:03:24Z)
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.