TinyJudge: Unverifiable Constraint Alignment via Lightweight Specialist Ensembles

• URL: http://arxiv.org/abs/2606.07520v1
• Date: Sun, 19 Apr 2026 06:02:15 GMT
• Title: TinyJudge: Unverifiable Constraint Alignment via Lightweight Specialist Ensembles
• Abstract summary: We propose TinyJudge, a framework that employs an ensemble of specialized tiny language models to provide rewards for soft constraints.<n>TinyJudge outperforms the baselines by $sim10%$ in average performance and $12%$ in reward precision.<n>Our work provides a scalable and robust path for aligning LLMs with unverifiable human instructions.
• Score: 28.896629356988736
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Instruction Following (IF) is a core capability of LLMs, requiring strict adherence to diverse constraints, ranging from verifiable ones (e.g., output length) to unverifiable ones (e.g., tone). Reinforcement learning with verifiable rewards has emerged as a paradigm for IF tasks, leveraging LLM-as-a-judge to assess unverifiable constraints. However, we empirically find that this approach remains a significant bottleneck, suffering from severe reward hacking and higher computational overhead. In this work, we first analyze the generalization capabilities of unverifiable constraints and discover that specific constraints exhibit distinct, high-generalization patterns. Motivated by this, we propose TinyJudge, a framework that employs an ensemble of specialized tiny language models ($\sim0.6B$) to provide rewards for soft constraints. By distilling expertise from frontier models into these tiny models, it achieves high-precision, lightweight evaluation. Extensive evaluations across five benchmarks demonstrate that TinyJudge outperforms the baselines by $\sim10\%$ in average performance and $12\%$ in reward precision. Crucially, it also achieves a $3\times$ speedup in total training time. Our work provides a scalable and robust path for aligning LLMs with unverifiable human instructions.

Related papers

• Theoretical Limits of Language Model Alignment [9.45142272392467]
  Language model (LM) alignment improves model outputs to reflect human preferences while preserving the capabilities of the base model.<n>The most common alignment approaches are (i) reinforcement learning, which maximizes the expected reward under a KL-divergence constraint, and (ii) best-of-$N$ alignment, which selects the highest-reward output among $N$ independent samples.<n>We characterize the information-theoretic limits of KL-regularized alignment by deriving the maximum achievable expected reward gain for a fixed KL-divergence budget.
  arXiv  Detail & Related papers  (2026-05-08T01:32:22Z)
• RVPO: Risk-Sensitive Alignment via Variance Regularization [13.192921543523283]
  We propose Reward-Variance Policy Optimization (RVPO), a risk-sensitive framework that penalizes inter-reward variance during advantage aggregation.<n>We show via Taylor expansion that a LogSumExp (SoftMin) operator effectively acts as a smooth variance penalty.<n>By preventing the model from neglecting difficult constraints to exploit easier objectives, RVPO improves overall scores on HealthBench.
  arXiv  Detail & Related papers  (2026-05-07T06:43:05Z)
• LLM Readiness Harness: Evaluation, Observability, and CI Gates for LLM/RAG Applications [51.56484100374058]
  We present a readiness harness for LLM and RAG applications that turns evaluation into a deployment decision workflow.<n>The system combines automated benchmarks, OpenTelemetry observability, and CI quality gates under a minimal API contract.<n>We evaluate the harness on ticket-routing and BEIR tasks with full Azure matrix coverage.
  arXiv  Detail & Related papers  (2026-03-28T18:03:32Z)
• Robust LLM Performance Certification via Constrained Maximum Likelihood Estimation [14.465019747508604]
  We propose a new, practical, and efficient approach to LLM failure rate estimation based on constrained maximum-likelihood estimation (MLE)<n>We validate our approach through a comprehensive empirical study, benchmarking it against state-of-the-art baselines like Prediction-Powered Inference (PPI)
  arXiv  Detail & Related papers  (2026-03-11T21:48:40Z)
• What's the Price of Monotonicity? A Multi-Dataset Benchmark of Monotone-Constrained Gradient Boosting for Credit PD [0.0]
  Financial institutions face a trade-off between predictive accuracy and interpretability when deploying machine learning models for credit risk.<n>Monotonicity constraints align model behavior with domain knowledge, but their performance cost is not well quantified.<n>This paper benchmarks monotone-constrained versus unconstrained gradient boosting models for credit probability of default across five public datasets and three libraries.
  arXiv  Detail & Related papers  (2025-12-14T22:18:05Z)
• IF-CRITIC: Towards a Fine-Grained LLM Critic for Instruction-Following Evaluation [87.38454788767545]
  We propose IF-CRITIC, an evaluation model for instruction following in Large Language Models.<n>With the scalable reward signals provided by IF-CRITIC, LLMs can achieve substantial performance gains in instruction-following optimization.
  arXiv  Detail & Related papers  (2025-11-02T17:06:49Z)
• Large-Scale Constraint Generation - Can LLMs Parse Hundreds of Constraints? [20.47853693801739]
  Large-Scale Constraint Generation (LSCG) is a new problem that evaluates whether Large Language Models can parse a large, fine-grained, generic list of constraints.<n>In Words Checker, we evaluate the impact of model characteristics (e.g., size, family) and steering techniques (e.g., Simple Prompt, Chain of Thought, Best of N) on performance.<n>We propose FoCusNet, a small and dedicated model that parses the original list of constraints into a smaller subset, helping the LLM focus on relevant constraints.
  arXiv  Detail & Related papers  (2025-09-28T21:55:53Z)
• A Fano-Style Accuracy Upper Bound for LLM Single-Pass Reasoning in Multi-Hop QA [65.38186593873313]
  Multi-Hop Question Answering (MHQA) requires integrating dispersed, interdependent evidence through sequential reasoning under noise.<n>We introduce a proof-of-concept multi-call framework for MHQA, InfoQA.<n>We construct a stringent and noise-rich benchmark to validate our theory and framework.
  arXiv  Detail & Related papers  (2025-09-25T14:11:57Z)
• Supervised Optimism Correction: Be Confident When LLMs Are Sure [91.7459076316849]
  We establish a novel theoretical connection between supervised fine-tuning and offline reinforcement learning.<n>We show that the widely used beam search method suffers from unacceptable over-optimism.<n>We propose Supervised Optimism Correction, which introduces a simple yet effective auxiliary loss for token-level $Q$-value estimations.
  arXiv  Detail & Related papers  (2025-04-10T07:50:03Z)
• Attribute Controlled Fine-tuning for Large Language Models: A Case Study on Detoxification [76.14641982122696]
  We propose a constraint learning schema for fine-tuning Large Language Models (LLMs) with attribute control. We show that our approach leads to an LLM that produces fewer inappropriate responses while achieving competitive performance on benchmarks and a toxicity detection task.
  arXiv  Detail & Related papers  (2024-10-07T23:38:58Z)
• FollowBench: A Multi-level Fine-grained Constraints Following Benchmark for Large Language Models [79.62191017182518]
  FollowBench is a benchmark for Fine-grained Constraints Following Benchmark for Large Language Models. We introduce a Multi-level mechanism that incrementally adds a single constraint to the initial instruction at each increased level. By evaluating 13 popular LLMs on FollowBench, we highlight the weaknesses of LLMs in instruction following and point towards potential avenues for future work.
  arXiv  Detail & Related papers  (2023-10-31T12:32:38Z)

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.