An official website of the United States government
Existing preference optimization objectives for language model alignment require additional hyperparameters that must be extensively tuned to achieve optimal performance, increasing both the complexity and time required for fine-tuning large language models. In this paper, we propose a simple yet effective hyperparameter-free preference optimization algorithm for alignment. We observe that promising performance can be achieved simply by optimizing inverse perplexity, which is calculated as the inverse of the exponentiated average log-likelihood of the chosen and rejected responses in the preference dataset. The resulting simple learning objective, SimPER, is easy to implement and eliminates the need for expensive hyperparameter tuning and a reference model, making it both computationally and memory efficient. Extensive experiments on widely used real-world benchmarks, including MT-Bench, AlpacaEval 2, and 10 key benchmarks of the Open LLM Leaderboard with 5 base models, demonstrate that SimPER consistently and significantly outperforms existing approaches—even without any hyperparameters or a reference model. For example, despite its simplicity, SimPER outperforms state-of-the-art methods by up to 5.7 points on AlpacaEval 2 and achieves the highest average ranking across 10 benchmarks on the Open LLM Leaderboard. The source code for SimPER is publicly available at: https://github.com/tengxiao1/SimPER.
more »
« less
This content will become publicly available on July 8, 2026
The Delta Learning Hypothesis: Preference Tuning on Weak Data can Yield Strong Gains
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.
more »
« less
- Award ID(s):
- 2505865
- PAR ID:
- 10631465
- Publisher / Repository:
- https://doi.org/10.48550/arXiv.2507.06187
- Date Published:
- ISSN:
- 2507.06187
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract MotivationQuality assessment (QA) of predicted protein tertiary structure models plays an important role in ranking and using them. With the recent development of deep learning end-to-end protein structure prediction techniques for generating highly confident tertiary structures for most proteins, it is important to explore corresponding QA strategies to evaluate and select the structural models predicted by them since these models have better quality and different properties than the models predicted by traditional tertiary structure prediction methods. ResultsWe develop EnQA, a novel graph-based 3D-equivariant neural network method that is equivariant to rotation and translation of 3D objects to estimate the accuracy of protein structural models by leveraging the structural features acquired from the state-of-the-art tertiary structure prediction method—AlphaFold2. We train and test the method on both traditional model datasets (e.g. the datasets of the Critical Assessment of Techniques for Protein Structure Prediction) and a new dataset of high-quality structural models predicted only by AlphaFold2 for the proteins whose experimental structures were released recently. Our approach achieves state-of-the-art performance on protein structural models predicted by both traditional protein structure prediction methods and the latest end-to-end deep learning method—AlphaFold2. It performs even better than the model QA scores provided by AlphaFold2 itself. The results illustrate that the 3D-equivariant graph neural network is a promising approach to the evaluation of protein structural models. Integrating AlphaFold2 features with other complementary sequence and structural features is important for improving protein model QA. Availability and implementationThe source code is available at https://github.com/BioinfoMachineLearning/EnQA. Supplementary informationSupplementary data are available at Bioinformatics online.more » « less
-
The weak-to-strong generalization phenomenon is the driver for important machine learning applications including highly data-efficient learning and, most recently, performing superalignment. While decades of research have resulted in numerous algorithms that produce strong empirical performance, understanding what aspects of data enable weak-to-strong generalization has been understudied. We propose a simple data-centric mechanism that characterizes weak-to-strong generalization: the overlap density. Intuitively, generalization tracks the number of points that contain overlaps, i.e., both easy patterns (learnable by a weak model) and challenging patterns (only learnable by a stronger model), as with such points, weak predictions can be used to learn challenging patterns by stronger models. We provide a practical overlap detection algorithm to find such points in datasets and leverage them to learn, among multiple sources of data, which to query when seeking to maximize overlap density and thereby enhance weak-to-strong generalization. We present a theoretical result showing that the generalization benefit is a function of the overlap density and a regret bound for our data selection algorithm. Empirically, we validate the mechanism and the overlap detection algorithm on a wide array of settings.more » « less
-
The weak-to-strong generalization phenomenon is the driver for important machine learning applications including highly data-efficient learning and, most recently, performing superalignment. While decades of research have resulted in numerous algorithms that produce strong empirical performance, understanding what aspects of data enable weak-to-strong generalization has been understudied. We propose a simple data-centric mechanism that characterizes weak-to-strong generalization: the overlap density. Intuitively, generalization tracks the number of points that contain overlaps, i.e., both easy patterns (learnable by a weak model) and challenging patterns (only learnable by a stronger model), as with such points, weak predictions can be used to learn challenging patterns by stronger models. We provide a practical overlap detection algorithm to find such points in datasets and leverage them to learn, among multiple sources of data, which to query when seeking to maximize overlap density and thereby enhance weak-to-strong generalization. We present a theoretical result showing that the generalization benefit is a function of the overlap density and a regret bound for our data selection algorithm. Empirically, we validate the mechanism and the overlap detection algorithm on a wide array of settings.more » « less
-
Varied approaches for aligning language models have been proposed, including supervised fine-tuning, RLHF, and direct optimization methods such as DPO. Although DPO has rapidly gained popularity due to its straightforward training process and competitive results, there is an open question of whether there remain practical advantages of using a discriminator, like a reward model, to evaluate responses. We propose D2PO, discriminator-guided DPO, an approach for the online setting where preferences are being collected throughout learning. As we collect gold preferences, we use these not only to train our policy, but to train a discriminative response evaluation model to silver-label even more synthetic data for policy training. We explore this approach across a set of diverse tasks, including a realistic chat setting, we find that our approach leads to higher-quality outputs compared to DPO with the same data budget, and greater efficiency in terms of preference data requirements. Furthermore, we show conditions under which silver labeling is most helpful: it is most effective when training the policy with DPO, outperforming traditional PPO, and benefits from maintaining a separate discriminator from the policy model.more » « less
