Evaluating Judges as Evaluators: The JETTS Benchmark of LLM-as-Judges as Test-Time Scaling Evaluators

Scaling test-time computation, or affording a generator large language model (LLM) extra compute during inference, typically employs the help of external non-generative evaluators (i.e., reward models). Concurrently, LLM-judges, models trained to generate evaluations and critiques (explanations) in natural language, are becoming increasingly popular in automatic evaluation. Despite judge empirical successes, their effectiveness as evaluators in test-time scaling settings is largely unknown. In this paper, we introduce the Judge Evaluation for Test-Time Scaling (JETTS) benchmark, which evaluates judge performance in three domains (math reasoning, code generation, and instruction following) under three task settings: response reranking, step-level beam search, and critique-based response refinement. We evaluate 10 different judge models (7B-70B parameters) for 8 different base generator models (6.7B-72B parameters). Our benchmark shows that while judges are competitive with outcome reward models in reranking, they are consistently worse than process reward models in beam search procedures. Furthermore, though unique to LLM-judges, their natural language critiques are currently ineffective in guiding the generator towards better responses.

A Survey of Frontiers in LLM Reasoning: Inference Scaling, Learning to Reason, and Agentic Systems

Reasoning is a fundamental cognitive process that enables logical inference, problem-solving, and decision-making. With the rapid advancement of large language models (LLMs), reasoning has emerged as a key capability that distinguishes advanced AI systems from conventional models that empower chatbots. In this survey, we categorize existing methods along two orthogonal dimensions: (1) Regimes, which define the stage at which reasoning is achieved (either at inference time or through dedicated training); and (2) Architectures, which determine the components involved in the reasoning process, distinguishing between standalone LLMs and agentic compound systems that incorporate external tools, and multi-agent collaborations. Within each dimension, we analyze two key perspectives: (1) Input level, which focuses on techniques that construct high-quality prompts that the LLM condition on; and (2) Output level, which methods that refine multiple sampled candidates to enhance reasoning quality. This categorization provides a systematic understanding of the evolving landscape of LLM reasoning, highlighting emerging trends such as the shift from inference-scaling to learning-to-reason (e.g., DeepSeek-R1), and the transition to agentic workflows (e.g., OpenAI Deep Research, Manus Agent). Additionally, we cover a broad spectrum of learning algorithms, from supervised fine-tuning to reinforcement learning such as PPO and GRPO, and the training of reasoners and verifiers. We also examine key designs of agentic workflows, from established patterns like generator-evaluator and LLM debate to recent innovations. ...

ChartQAPro: A More Diverse and Challenging Benchmark for Chart Question Answering

Charts are ubiquitous, as people often use them to analyze data, answer questions, and discover critical insights. However, performing complex analytical tasks with charts requires significant perceptual and cognitive effort. Chart Question Answering (CQA) systems automate this process by enabling models to interpret and reason with visual representations of data. However, existing benchmarks like ChartQA lack real-world diversity and have recently shown performance saturation with modern large vision-language models (LVLMs). To address these limitations, we introduce ChartQAPro, a new benchmark that includes 1,341 charts from 157 diverse sources, spanning various chart types, including infographics and dashboards, and featuring 1,948 questions in various types, such as multiple-choice, conversational, hypothetical, and unanswerable questions, to better reflect real-world challenges. Our evaluations with 21 models show a substantial performance drop for LVLMs on ChartQAPro; e.g., Claude Sonnet 3.5 scores 90.5% on ChartQA but only 55.81% on ChartQAPro, underscoring the complexity of chart reasoning. We complement our findings with detailed error analyses and ablation studies, identifying key challenges and opportunities for advancing LVLMs in chart understanding and reasoning. We release ChartQAPro at https://github.com/vis-nlp/ChartQAPro.

Adaptation of Large Language Models

This tutorial on adaptation of LLMs is designed to address the growing demand for models that go beyond the static capabilities of generic LLMs by providing an overview of dynamic, domain-specific, and task-adaptive LLM adaptation techniques. While general LLMs have demonstrated strong generalization across a variety of tasks, they often struggle to perform well in specialized domains such as finance, healthcare, and code generation for underrepresented languages. Additionally, their static nature limits their ability to evolve with the changing world, and they are often extremely large in size, making them impractical and costly to deploy at scale. As a result, the adaptation of LLMs has drawn much attention since the birth of LLMs and is of core importance, both for industry, which focuses on serving its targeted users, and academia, which can greatly benefit from small but powerful LLMs. To address this gap, this tutorial aims to provide an overview of the LLM adaptation techniques. We start with an introduction to LLM adaptation, from both the data perspective and the model perspective. We then emphasize how the evaluation metrics and benchmarks are different from other techniques. After establishing the problems, we explore various adaptation techniques. We categorize adaptation techniques into two main families. The first is parametric knowledge adaptation, which focuses on updating the parametric knowledge within LLMs. Additionally, we will discuss real-time adaptation techniques, including model editing, which allows LLMs to be updated dynamically in production environments. The second kind of adaptation is semi-parametric knowledge adaptation, where the goal is to update LLM parameters to better leverage external knowledge or tools through techniques like retrieval-augmented generation (RAG) and agent-based systems.

Does Context Matter? ContextualJudgeBench for Evaluating LLM-based Judges in Contextual Settings

The large language model (LLM)-as-judge paradigm has been used to meet the demand for a cheap, reliable, and fast evaluation of model outputs during AI system development and post-deployment monitoring. While judge models -- LLMs finetuned to specialize in assessing and critiquing model outputs -- have been touted as general purpose evaluators, they are typically evaluated only on non-contextual scenarios, such as instruction following. The omission of contextual settings -- those where external information is used as context to generate an output -- is surprising given the increasing prevalence of retrieval-augmented generation (RAG) and summarization use cases. Contextual assessment is uniquely challenging, as evaluation often depends on practitioner priorities, leading to conditional evaluation criteria (e.g., comparing responses based on factuality and then considering completeness if they are equally factual). To address the gap, we propose ContextualJudgeBench, a judge benchmark with 2,000 challenging response pairs across eight splits inspired by real-world contextual evaluation scenarios. We build our benchmark with a multi-pronged data construction pipeline that leverages both existing human annotations and model-based perturbations. Our comprehensive study across 11 judge models and 9 general purpose models, reveals that the contextual information and its assessment criteria present a significant challenge to even state-of-the-art models. For example, OpenAI's o1, the best-performing model, barely reaches 55% consistent accuracy.

Multi$^2$: Multi-Agent Test-Time Scalable Framework for Multi-Document Processing

Recent advances in test-time scaling have shown promising results in improving Large Language Models (LLMs) performance through strategic computation allocation during inference. While this approach has demonstrated strong performance improvements in logical and mathematical reasoning tasks, its application to natural language generation (NLG), especially summarization, has yet to be explored. Multi-Document Summarization (MDS) is a challenging task that focuses on extracting and synthesizing useful information from multiple lengthy documents. Unlike reasoning tasks, MDS requires a more nuanced approach to prompt design and ensemble, as there is no "best" prompt to satisfy diverse summarization requirements. To address this, we propose a novel framework that leverages inference-time scaling for this task. Precisely, we take prompt ensemble approach by leveraging various prompt to first generate candidate summaries and then ensemble them with an aggregator to produce a refined summary. We also introduce two new evaluation metrics: Consistency-Aware Preference (CAP) score and LLM Atom-Content-Unit (ACU) score, to enhance LLM's contextual understanding while mitigating its positional bias. Extensive experiments demonstrate the effectiveness of our approach in improving summary quality while identifying and analyzing the scaling boundaries in summarization tasks.

Why Vision Language Models Struggle with Visual Arithmetic? Towards Enhanced Chart and Geometry Understanding

Vision Language Models (VLMs) have achieved remarkable progress in multimodal tasks, yet they often struggle with visual arithmetic, seemingly simple capabilities like object counting or length comparison, which are essential for relevant complex tasks like chart understanding and geometric reasoning. In this work, we first investigate the root causes of this deficiency through a suite of probing tasks focusing on basic visual arithmetic. Our analysis reveals that while pre-trained vision encoders typically capture sufficient information, the text decoder often fails to decode it correctly for arithmetic reasoning. To address this, we propose CogAlign, a novel post-training strategy inspired by Piaget's theory of cognitive development. CogAlign trains VLMs to recognize invariant properties under visual transformations. We demonstrate that this approach significantly improves the performance of three diverse VLMs on our proposed probing tasks. Furthermore, CogAlign enhances performance by an average of 4.6% on CHOCOLATE and 2.9% on MATH-VISION, outperforming or matching supervised fine-tuning methods while requiring only 60% less training data. These results highlight the effectiveness and generalizability of CogAlign in improving fundamental visual arithmetic capabilities and their transfer to downstream tasks.

Demystifying Domain-adaptive Post-training for Financial LLMs

Domain-adaptive post-training of large language models (LLMs) has emerged as a promising approach for specialized domains such as medicine and finance. However, significant challenges remain in identifying optimal adaptation criteria and training strategies across varying data and model configurations. To address these challenges, we introduce FINDAP, a systematic and fine-grained investigation into domain-adaptive post-training of LLMs for the finance domain. Our approach begins by identifying the core capabilities required for the target domain and designing a comprehensive evaluation suite aligned with these needs. We then analyze the effectiveness of key post-training stages, including continual pretraining, instruction tuning, and preference alignment. Building on these insights, we propose an effective training recipe centered on a novel preference data distillation method, which leverages process signals from a generative reward model. The resulting model, Llama-Fin, achieves state-of-the-art performance across a wide range of financial tasks. Our analysis also highlights how each post-training stage contributes to distinct capabilities, uncovering specific challenges and effective solutions, providing valuable insights for domain adaptation of LLMs. Project page: https://github.com/SalesforceAIResearch/FinDap

StructTest: Benchmarking LLMs' Reasoning through Compositional Structured Outputs

The rapid development of large language models (LLMs) necessitates robust, unbiased, and scalable methods for evaluating their capabilities. However, human annotations are expensive to scale, model-based evaluations are prone to biases in answer style, while target-answer-based benchmarks are vulnerable to data contamination and cheating. To address these limitations, we propose StructTest, a novel benchmark that evaluates LLMs on their ability to produce compositionally specified structured outputs as an unbiased, cheap-to-run and difficult-to-cheat measure. The evaluation is done deterministically by a rule-based evaluator, which can be easily extended to new tasks. By testing structured outputs across diverse task domains -- including Summarization, Code, HTML and Math -- we demonstrate that StructTest serves as a good proxy for general reasoning abilities, as producing structured outputs often requires internal logical reasoning. We believe that StructTest offers a critical, complementary approach to objective and robust model evaluation.

Preference Optimization for Reasoning with Pseudo Feedback

Preference optimization techniques, such as Direct Preference Optimization (DPO), are frequently employed to enhance the reasoning capabilities of large language models (LLMs) in domains like mathematical reasoning and coding, typically following supervised fine-tuning. These methods rely on high-quality labels for reasoning tasks to generate preference pairs; however, the availability of reasoning datasets with human-verified labels is limited. In this study, we introduce a novel approach to generate pseudo feedback for reasoning tasks by framing the labeling of solutions to reason problems as an evaluation against associated test cases. We explore two forms of pseudo feedback based on test cases: one generated by frontier LLMs and the other by extending self-consistency to multi-test-case. We conduct experiments on both mathematical reasoning and coding tasks using pseudo feedback for preference optimization, and observe improvements across both tasks. Specifically, using Mathstral-7B as our base model, we improve MATH results from 58.3 to 68.6, surpassing both NuminaMath-72B and GPT-4-Turbo-1106-preview. In GSM8K and College Math, our scores increase from 85.6 to 90.3 and from 34.3 to 42.3, respectively. Building on Deepseek-coder-7B-v1.5, we achieve a score of 24.6 on LiveCodeBench (from 21.1), surpassing Claude-3-Haiku.