Less Data, Better Alignment: Data-Centric Multi-Evaluator Agreement for Preference Optimization

• URL: http://arxiv.org/abs/2607.25136v1
• Date: Mon, 27 Jul 2026 23:05:12 GMT
• Title: Less Data, Better Alignment: Data-Centric Multi-Evaluator Agreement for Preference Optimization
• Abstract summary: DMAPO generates candidate responses from the target policy, evaluates helpfulness, factuality, and conciseness with rubric-specialized evaluators.<n>This procedure accepts 1,871 of 54,236 Mistral-7B candidates (3.45%)<n>KTO trained on this set reaches 7.50 on MT-Bench, 95.5% length-controlled win rate against a text-davinci-003 reference, and 57.3% IFEval prompt accuracy.
• Score: 20.147316178118732
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Research on preference optimization often varies the training objective while holding the data fixed. We instead ask whether a small, high-confidence set of on-policy responses can provide a reliable learning signal. Our method, DMAPO (Data-centric Multi-evaluator Agreement for Preference Optimization), generates candidate responses from the target policy, evaluates helpfulness, factuality, and conciseness with rubric-specialized evaluators, applies a process-critic correction, and retains only high-consensus desirable or undesirable examples. This procedure accepts 1,871 of 54,236 Mistral-7B candidates (3.45%). KTO trained on this set reaches 7.50 on MT-Bench, 95.5% length-controlled win rate against a text-davinci-003 reference, and 57.3% IFEval prompt accuracy. Independent pairwise evaluation also favors DMAPO over SimPO: GPT-4o yields a net win rate of 23.3 points on 129 held-out prompts and 24.0 points on 200 out-of-distribution LMSYS-Chat prompts; Claude Opus 4.7 yields 24.1 points on the held-out set. Changing the evaluator model or rubric alters the selected examples but has little effect on downstream performance. A second-backbone study yields a similar 3.41% acceptance rate, although its performance gains are more modest. Across these experiments, consensus filtering offers a data-efficient route to preference optimization for general instructions, at the cost of additional curation compute and dependence on evaluator judgments.

Related papers

• Do Agent Optimizers Compound? A Continual-Learning Evaluation on Terminal-Bench 2.0 [53.71882250666667]
  Most reported gains from agent-optimization methods are one-shot.<n>This does not test the setting that matters for deployed agents.<n>We compare three approaches to agent-harness optimization.
  arXiv  Detail & Related papers  (2026-07-15T16:36:04Z)
• Visual Preference Optimization with Rubric Rewards [30.826907231502663]
  We propose rDPO, a preference optimization framework based on instance-specific rubrics.<n>On public reward modeling benchmarks, rubric-based prompting massively improves a 30B-A3B judge and brings it close to GPT-5.4.<n>When evaluating scalability on a comprehensive benchmark, rDPO achieves 61.01, markedly outperforming the style-constrained baseline (52.36) and surpassing the 59.48 base model.
  arXiv  Detail & Related papers  (2026-04-14T17:58:22Z)
• Select Smarter, Not More: Prompt-Aware Evaluation Scheduling with Submodular Guarantees [28.54776477263591]
  We present Prompt-Aware Online Evaluation Scheduling (POES)<n>POES integrates an IRT-based discrimination utility, a facility-location coverage term, and switching-cost-aware warm-start swaps into a unified objective.<n>POES achieves the highest overall average accuracy (6.2 percent improvement over the best baseline) with negligible token overhead.
  arXiv  Detail & Related papers  (2026-04-13T11:31:04Z)
• Multi-Turn Reinforcement Learning for Tool-Calling Agents with Iterative Reward Calibration [0.0]
  Training tool-calling agents with reinforcement learning on multi-turn tasks remains challenging.<n>We present the first application of MT-GRPO and GTPO for training a tool-calling agent on realistic customer service tasks.
  arXiv  Detail & Related papers  (2026-04-03T08:36:03Z)
• Reverse Preference Optimization for Complex Instruction Following [61.39734201711077]
  We propose a simple yet effective method called Reverse Preference Optimization (RPO)<n>It mitigates noise in preference pairs by dynamically reversing the constraints within the instruction to ensure the chosen response is perfect.<n>RPO scales effectively across model sizes, with the 70B RPO model surpassing GPT-4o.
  arXiv  Detail & Related papers  (2025-05-28T09:44:27Z)
• ELSPR: Evaluator LLM Training Data Self-Purification on Non-Transitive Preferences via Tournament Graph Reconstruction [25.85736569130897]
  Pairwise evaluation of large language models (LLMs) has become the dominant paradigm for benchmarking open-ended tasks.<n>We show that this critical issue stems largely from low-quality data that contains inherently ambiguous preference pairs.<n>We propose ELSPR, a principled graph-theoretic framework that models pairwise preferences as tournament graphs.
  arXiv  Detail & Related papers  (2025-05-23T10:00:03Z)
• R.I.P.: Better Models by Survival of the Fittest Prompts [51.2293437372642]
  We introduce a method for evaluating data integrity based on the assumption that low-quality input prompts result in high variance and low quality responses.<n>This is achieved by measuring the rejected response quality and the reward gap between the chosen and rejected preference pair.
  arXiv  Detail & Related papers  (2025-01-30T18:50:25Z)
• Preference Optimization for Reasoning with Pseudo Feedback [100.62603571434167]
  We introduce a novel approach to generate pseudo feedback for reasoning tasks by framing the labeling of solutions as an evaluation against associated test cases.<n>We conduct experiments on both mathematical reasoning and coding tasks using pseudo feedback for preference optimization, and observe improvements across both tasks.
  arXiv  Detail & Related papers  (2024-11-25T12:44:02Z)
• Reward-Augmented Data Enhances Direct Preference Alignment of LLMs [63.32585910975191]
  We introduce reward-conditioned Large Language Models (LLMs) that learn from the entire spectrum of response quality within the dataset.<n>We show that our approach consistently boosts DPO by a considerable margin.<n>Our method not only maximizes the utility of preference data but also mitigates the issue of unlearning, demonstrating its broad effectiveness beyond mere data expansion.
  arXiv  Detail & Related papers  (2024-10-10T16:01:51Z)
• Anchored Preference Optimization and Contrastive Revisions: Addressing Underspecification in Alignment [57.03947082589616]
  Large Language Models (LLMs) are often aligned using contrastive alignment objectives and preference pair datasets. We study this and find that preference data gives a better learning signal when the underlying responses are contrastive. We introduce Contrastive Learning from AI Revisions (CLAIR), a data-creation method which leads to more contrastive preference pairs. Our best model, trained on 32K CLAIR preferences with APO, improves Llama-3-8B-Instruct by 7.65%, closing the gap with GPT4-turbo by 45%.
  arXiv  Detail & Related papers  (2024-08-12T16:24:51Z)
• 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), utilizes iterative policy updates to provably approximate the Nash equilibrium.
  arXiv  Detail & Related papers  (2024-05-01T17:59:20Z)

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.