Getting More Juice Out of the SFT Data: Reward Learning from Human Demonstration Improves SFT for LLM Alignment

• URL: http://arxiv.org/abs/2405.17888v2
• Date: Wed, 29 May 2024 13:33:33 GMT
• Title: Getting More Juice Out of the SFT Data: Reward Learning from Human Demonstration Improves SFT for LLM Alignment
• Authors: Jiaxiang Li, Siliang Zeng, Hoi-To Wai, Chenliang Li, Alfredo Garcia, Mingyi Hong,
• Abstract summary: State-of-the-art techniques such as Reinforcement Learning from Human Feedback (RLHF) often consist of two stages. 1) supervised fine-tuning (SFT), where the model is fine-tuned by learning from human demonstration data. 2) Preference learning, where preference data is used to learn a reward model, which is in turn used by a reinforcement learning step to fine-tune the model.
• Score: 65.15914284008973
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Aligning human preference and value is an important requirement for contemporary foundation models. State-of-the-art techniques such as Reinforcement Learning from Human Feedback (RLHF) often consist of two stages: 1) supervised fine-tuning (SFT), where the model is fine-tuned by learning from human demonstration data; 2) Preference learning, where preference data is used to learn a reward model, which is in turn used by a reinforcement learning (RL) step to fine-tune the model. Such reward model serves as a proxy to human preference, and it is critical to guide the RL step towards improving the model quality. In this work, we argue that the SFT stage significantly benefits from learning a reward model as well. Instead of using the human demonstration data directly via supervised learning, we propose to leverage an Inverse Reinforcement Learning (IRL) technique to (explicitly or implicitly) build an reward model, while learning the policy model. This approach leads to new SFT algorithms that are not only efficient to implement, but also promote the ability to distinguish between the preferred and non-preferred continuations. Moreover, we identify a connection between the proposed IRL based approach, and certain self-play approach proposed recently, and showed that self-play is a special case of modeling a reward-learning agent. Theoretically, we show that the proposed algorithms converge to the stationary solutions of the IRL problem. Empirically, we align 1B and 7B models using proposed methods and evaluate them on a reward benchmark model and the HuggingFace Open LLM Leaderboard. The proposed methods show significant performance improvement over existing SFT approaches. Our results indicate that it is beneficial to explicitly or implicitly leverage reward learning throughout the entire alignment process.

Related papers

• Weak-to-Strong Extrapolation Expedites Alignment [135.12769233630362]
  We propose a method called ExPO to boost models' alignment with human preference. We demonstrate that ExPO consistently improves off-the-shelf DPO/RLHF models. We shed light on the essence of ExPO amplifying the reward signal learned during alignment training.
  arXiv  Detail & Related papers  (2024-04-25T17:39:50Z)
• Secrets of RLHF in Large Language Models Part II: Reward Modeling [134.97964938009588]
  We introduce a series of novel methods to mitigate the influence of incorrect and ambiguous preferences in the dataset. We also introduce contrastive learning to enhance the ability of reward models to distinguish between chosen and rejected responses.
  arXiv  Detail & Related papers  (2024-01-11T17:56:59Z)
• SALMON: Self-Alignment with Instructable Reward Models [80.83323636730341]
  This paper presents a novel approach, namely SALMON, to align base language models with minimal human supervision. We develop an AI assistant named Dromedary-2 with only 6 exemplars for in-context learning and 31 human-defined principles.
  arXiv  Detail & Related papers  (2023-10-09T17:56:53Z)
• Direct Preference-based Policy Optimization without Reward Modeling [25.230992130108767]
  Preference-based reinforcement learning (PbRL) is an approach that enables RL agents to learn from preference. We propose a PbRL algorithm that directly learns from preference without requiring any reward modeling. We show that our algorithm surpasses offline RL methods that learn with ground-truth reward information.
  arXiv  Detail & Related papers  (2023-01-30T12:51:13Z)
• Reward Uncertainty for Exploration in Preference-based Reinforcement Learning [88.34958680436552]
  We present an exploration method specifically for preference-based reinforcement learning algorithms. Our main idea is to design an intrinsic reward by measuring the novelty based on learned reward. Our experiments show that exploration bonus from uncertainty in learned reward improves both feedback- and sample-efficiency of preference-based RL algorithms.
  arXiv  Detail & Related papers  (2022-05-24T23:22:10Z)
• Model-Augmented Q-learning [112.86795579978802]
  We propose a MFRL framework that is augmented with the components of model-based RL. Specifically, we propose to estimate not only the $Q$-values but also both the transition and the reward with a shared network. We show that the proposed scheme, called Model-augmented $Q$-learning (MQL), obtains a policy-invariant solution which is identical to the solution obtained by learning with true reward.
  arXiv  Detail & Related papers  (2021-02-07T17:56:50Z)
• Model Embedding Model-Based Reinforcement Learning [4.566180616886624]
  Model-based reinforcement learning (MBRL) has shown its advantages in sample-efficiency over model-free reinforcement learning (MFRL) Despite the impressive results it achieves, it still faces a trade-off between the ease of data generation and model bias. We propose a simple and elegant model-embedding model-based reinforcement learning (MEMB) algorithm in the framework of the probabilistic reinforcement learning.
  arXiv  Detail & Related papers  (2020-06-16T15:10:28Z)

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.