Investigating on RLHF methodology

• URL: http://arxiv.org/abs/2410.01789v1
• Date: Wed, 2 Oct 2024 17:46:22 GMT
• Title: Investigating on RLHF methodology
• Authors: Alexey Kutalev, Sergei Markoff,
• Abstract summary: We discuss the features of training a Preference Model, which simulates human preferences, and the methods and details we found essential for achieving the best results. We also discuss using Reinforcement Learning to fine-tune Large Language Models and describe the challenges we faced and the ways to overcome them.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: In this article, we investigate the alignment of Large Language Models according to human preferences. We discuss the features of training a Preference Model, which simulates human preferences, and the methods and details we found essential for achieving the best results. We also discuss using Reinforcement Learning to fine-tune Large Language Models and describe the challenges we faced and the ways to overcome them. Additionally, we present our experience with the Direct Preference Optimization method, which enables us to align a Large Language Model with human preferences without creating a separate Preference Model. As our contribution, we introduce the approach for collecting a preference dataset through perplexity filtering, which makes the process of creating such a dataset for a specific Language Model much easier and more cost-effective.

Related papers

• ComPO: Community Preferences for Language Model Personalization [122.54846260663922]
  ComPO is a method to personalize preference optimization in language models. We collect and release ComPRed, a question answering dataset with community-level preferences from Reddit.
  arXiv  Detail & Related papers  (2024-10-21T14:02:40Z)
• MetaAlign: Align Large Language Models with Diverse Preferences during Inference Time [50.41806216615488]
  Large Language Models (LLMs) acquire extensive knowledge and remarkable abilities from extensive text corpora. To make LLMs more usable, aligning them with human preferences is essential. We propose an effective method, textbf MetaAlign, which aims to help LLMs dynamically align with various explicit or implicit preferences specified at inference time.
  arXiv  Detail & Related papers  (2024-10-18T05:31:13Z)
• Unsupervised Human Preference Learning [7.959043497459107]
  Large language models demonstrate impressive reasoning abilities but struggle to provide personalized content. Existing methods, such as in-context learning and parameter-efficient fine-tuning, fall short in capturing the complexity of human preferences. We propose a novel approach utilizing small parameter models as preference agents to generate natural language rules that guide a larger, pre-trained model.
  arXiv  Detail & Related papers  (2024-09-30T17:51:01Z)
• Improving Context-Aware Preference Modeling for Language Models [62.32080105403915]
  We consider the two-step preference modeling procedure that first resolves the under-specification by selecting a context, and then evaluates preference with respect to the chosen context. We contribute context-conditioned preference datasets and experiments that investigate the ability of language models to evaluate context-specific preference.
  arXiv  Detail & Related papers  (2024-07-20T16:05:17Z)
• Language Representations Can be What Recommenders Need: Findings and Potentials [57.90679739598295]
  We show that item representations, when linearly mapped from advanced LM representations, yield superior recommendation performance. This outcome suggests the possible homomorphism between the advanced language representation space and an effective item representation space for recommendation. Our findings highlight the connection between language modeling and behavior modeling, which can inspire both natural language processing and recommender system communities.
  arXiv  Detail & Related papers  (2024-07-07T17:05:24Z)
• Aligning Large Language Models with Self-generated Preference Data [72.99676237703099]
  We propose a new framework that boosts the alignment of large language models (LLMs) with human preferences. Our key idea is leveraging the human prior knowledge within the small (seed) data. We introduce a noise-aware preference learning algorithm to mitigate the risk of low quality within generated preference data.
  arXiv  Detail & Related papers  (2024-06-06T18:01:02Z)
• Reuse Your Rewards: Reward Model Transfer for Zero-Shot Cross-Lingual Alignment [39.94156255629528]
  We evaluate a simple approach for zero-shot cross-lingual alignment. Cross-lingually aligned models are preferred by humans over unaligned models. A different-language reward model sometimes yields better aligned models than a same-language reward model.
  arXiv  Detail & Related papers  (2024-04-18T16:52:36Z)
• Optimizing Language Models for Human Preferences is a Causal Inference Problem [41.59906798328058]
  We present an initial exploration of language model optimization for human preferences from direct outcome datasets. We first propose that language model optimization should be viewed as a causal problem to ensure that the model correctly learns the relationship between the text and the outcome. We extend CPO with doubly robust CPO, which reduces the variance of the surrogate objective while retaining provably strong guarantees on bias.
  arXiv  Detail & Related papers  (2024-02-22T21:36:07Z)
• Active Preference Learning for Large Language Models [12.093302163058436]
  We develop an active learning strategy for DPO to make better use of preference labels. We propose a practical acquisition function for prompt/completion pairs based on the predictive entropy of the language model. We demonstrate how our approach improves both the rate of learning and final performance of fine-tuning on pairwise preference data.
  arXiv  Detail & Related papers  (2024-02-12T23:09:00Z)
• ULMA: Unified Language Model Alignment with Human Demonstration and Point-wise Preference [16.73260713938154]
  A typical alignment procedure consists of supervised fine-tuning and preference learning. We introduce Point-wise Direct Preference Optimization, a novel preference learning method designed to harness point-wise feedback effectively. Our work also uncovers a novel connection between supervised fine-tuning and point-wise preference learning, culminating in Unified Language Model Alignment.
  arXiv  Detail & Related papers  (2023-12-05T07:52:12Z)
• Chain of Hindsight Aligns Language Models with Feedback [62.68665658130472]
  We propose a novel technique, Chain of Hindsight, that is easy to optimize and can learn from any form of feedback, regardless of its polarity. We convert all types of feedback into sequences of sentences, which are then used to fine-tune the model. By doing so, the model is trained to generate outputs based on feedback, while learning to identify and correct negative attributes or errors.
  arXiv  Detail & Related papers  (2023-02-06T10:28:16Z)

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.