The Delta Learning Hypothesis: Preference Tuning on Weak Data can Yield Strong Gains

• URL: http://arxiv.org/abs/2507.06187v1
• Date: Tue, 08 Jul 2025 17:14:44 GMT
• Title: The Delta Learning Hypothesis: Preference Tuning on Weak Data can Yield Strong Gains
• Authors: Scott Geng, Hamish Ivison, Chun-Liang Li, Maarten Sap, Jerry Li, Ranjay Krishna, Pang Wei Koh,
• Abstract summary: We show that paired preference data consisting of individually weak data points can enable gains beyond the strength of each individual data point.<n>Our work shows that models can learn surprisingly well from paired data that might typically be considered weak.
• Score: 50.66245575710432
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Improvements in language models are often driven by improving the quality of the data we train them on, which can be limiting when strong supervision is scarce. In this work, we show that paired preference data consisting of individually weak data points can enable gains beyond the strength of each individual data point. We formulate the delta learning hypothesis to explain this phenomenon, positing that the relative quality delta between points suffices to drive learning via preference tuning--even when supervised finetuning on the weak data hurts. We validate our hypothesis in controlled experiments and at scale, where we post-train 8B models on preference data generated by pairing a small 3B model's responses with outputs from an even smaller 1.5B model to create a meaningful delta. Strikingly, on a standard 11-benchmark evaluation suite (MATH, MMLU, etc.), our simple recipe matches the performance of Tulu 3, a state-of-the-art open model tuned from the same base model while relying on much stronger supervisors (e.g., GPT-4o). Thus, delta learning enables simpler and cheaper open recipes for state-of-the-art post-training. To better understand delta learning, we prove in logistic regression that the performance gap between two weak teacher models provides useful signal for improving a stronger student. Overall, our work shows that models can learn surprisingly well from paired data that might typically be considered weak.

Related papers

• SPaRFT: Self-Paced Reinforcement Fine-Tuning for Large Language Models [51.74498855100541]
  Large language models (LLMs) have shown strong reasoning capabilities when fine-tuned with reinforcement learning (RL)<n>We propose textbfSPaRFT, a self-paced learning framework that enables efficient learning based on the capability of the model being trained.
  arXiv  Detail & Related papers  (2025-08-07T03:50:48Z)
• Forgetting: A New Mechanism Towards Better Large Language Model Fine-tuning [53.398270878295754]
  Supervised fine-tuning (SFT) plays a critical role for pretrained large language models (LLMs)<n>We suggest categorizing tokens within each corpus into two parts -- positive and negative tokens -- based on whether they are useful to improve model performance.<n>We conduct experiments on well-established benchmarks, finding that this forgetting mechanism not only improves overall model performance and also facilitate more diverse model responses.
  arXiv  Detail & Related papers  (2025-08-06T11:22:23Z)
• EpiCoDe: Boosting Model Performance Beyond Training with Extrapolation and Contrastive Decoding [50.29046178980637]
  EpiCoDe is a method that boosts model performance in data-scarcity scenarios without extra training.<n>We show that EpiCoDe consistently outperforms existing methods with significant and robust improvement.
  arXiv  Detail & Related papers  (2025-06-04T02:11:54Z)
• Smaller, Weaker, Yet Better: Training LLM Reasoners via Compute-Optimal Sampling [18.23215026159686]
  Training on high-quality synthetic data from strong language models (LMs) is a common strategy to improve the reasoning performance of LMs. We investigate the trade-offs between generating synthetic data using a stronger but more expensive (SE) model versus a weaker but cheaper (WC) model.
  arXiv  Detail & Related papers  (2024-08-29T17:32:35Z)
• Weak Reward Model Transforms Generative Models into Robust Causal Event Extraction Systems [17.10762463903638]
  We train evaluation models to approximate human evaluation, achieving high agreement. We propose a weak-to-strong supervision method that uses a fraction of the annotated data to train an evaluation model.
  arXiv  Detail & Related papers  (2024-06-26T10:48:14Z)
• Fantastic Gains and Where to Find Them: On the Existence and Prospect of General Knowledge Transfer between Any Pretrained Model [74.62272538148245]
  We show that for arbitrary pairings of pretrained models, one model extracts significant data context unavailable in the other. We investigate if it is possible to transfer such "complementary" knowledge from one model to another without performance degradation.
  arXiv  Detail & Related papers  (2023-10-26T17:59:46Z)
• Alleviating the Effect of Data Imbalance on Adversarial Training [26.36714114672729]
  We study adversarial training on datasets that obey the long-tailed distribution. We propose a new adversarial training framework -- Re-balancing Adversarial Training (REAT)
  arXiv  Detail & Related papers  (2023-07-14T07:01:48Z)
• Towards a robust and reliable deep learning approach for detection of compact binary mergers in gravitational wave data [0.0]
  We develop a deep learning model stage-wise and work towards improving its robustness and reliability. We retrain the model in a novel framework involving a generative adversarial network (GAN) Although absolute robustness is practically impossible to achieve, we demonstrate some fundamental improvements earned through such training.
  arXiv  Detail & Related papers  (2023-06-20T18:00:05Z)
• Contrastive Model Inversion for Data-Free Knowledge Distillation [60.08025054715192]
  We propose Contrastive Model Inversion, where the data diversity is explicitly modeled as an optimizable objective. Our main observation is that, under the constraint of the same amount of data, higher data diversity usually indicates stronger instance discrimination. Experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet demonstrate that CMI achieves significantly superior performance when the generated data are used for knowledge distillation.
  arXiv  Detail & Related papers  (2021-05-18T15:13:00Z)
• MixKD: Towards Efficient Distillation of Large-scale Language Models [129.73786264834894]
  We propose MixKD, a data-agnostic distillation framework, to endow the resulting model with stronger generalization ability. We prove from a theoretical perspective that under reasonable conditions MixKD gives rise to a smaller gap between the error and the empirical error. Experiments under a limited-data setting and ablation studies further demonstrate the advantages of the proposed approach.
  arXiv  Detail & Related papers  (2020-11-01T18:47:51Z)

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.