arXiv subjects

Zhifang Sui

Publications and source records attributed to Zhifang Sui.

At least 37 records · Page 2Linked to original sources

arXiv2025

Enhancing Reliability across Short and Long-Form QA via Reinforcement Learning

While reinforcement learning has unlocked unprecedented complex reasoning in large language models, it has also amplified their propensity for hallucination, creating a critical trade-off between capability and reliability. This work confronts this challenge by introducing a targeted RL framework designed to mitigate both intrinsic and extrinsic hallucinations across short and long-form question answering. We address extrinsic hallucinations (flawed internal knowledge) by creating a novel training set from open-ended conversions of TriviaQA. Concurrently, we tackle intrinsic hallucinations (unfaithfulness to context) by leveraging long-form texts from FineWeb in a fact-grounding reward scheme. To further bolster reliability, our framework explicitly rewards the model for refusing to answer unanswerable questions, thereby cultivating crucial cautiousness. Extensive experiments demonstrate that our methodology yields significant performance gains across a diverse suite of benchmarks, substantially reducing both hallucination types. Ultimately, this research contributes a practical framework for resolving the critical tension between advanced reasoning and factual trustworthiness, paving the way for more capable and reliable large language models.

cs.CL↗

arXiv2025

Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routers

Reinforcement learning (RL) has emerged as a crucial approach for enhancing the capabilities of large language models. However, in Mixture-of-Experts (MoE) models, the routing mechanism often introduces instability, even leading to catastrophic RL training collapse. We analyze the training-inference consistency of MoE models and identify a notable discrepancy in routing behaviors between the two phases. Moreover, even under identical conditions, the routing framework can yield divergent expert selections across repeated forward passes. To address this foundational inconsistency, we propose Rollout Routing Replay (R3), a method that records routing distributions from the inference engine and replays them during training. R3 significantly reduces training-inference policy KL divergence and mitigates extreme discrepancies without compromising training speed. Extensive experiments on various settings confirm that R3 succeeds in stabilizing RL training, preventing collapse and outperforming methods such as GSPO and TIS. We believe this work can offer a new solution for stabilizing RL in MoE models.

cs.CL↗

arXiv2025

LLM-REVal: Can We Trust LLM Reviewers Yet?

The rapid advancement of large language models (LLMs) has inspired researchers to integrate them extensively into the academic workflow, potentially reshaping how research is practiced and reviewed. While previous studies highlight the potential of LLMs in supporting research and peer review, their dual roles in the academic workflow and the complex interplay between research and review bring new risks that remain largely underexplored. In this study, we focus on how the deep integration of LLMs into both peer-review and research processes may influence scholarly fairness, examining the potential risks of using LLMs as reviewers by simulation. This simulation incorporates a research agent, which generates papers and revises, alongside a review agent, which assesses the submissions. Based on the simulation results, we conduct human annotations and identify pronounced misalignment between LLM-based reviews and human judgments: (1) LLM reviewers systematically inflate scores for LLM-authored papers, assigning them markedly higher scores than human-authored ones; (2) LLM reviewers persistently underrate human-authored papers with critical statements (e.g., risk, fairness), even after multiple revisions. Our analysis reveals that these stem from two primary biases in LLM reviewers: a linguistic feature bias favoring LLM-generated writing styles, and an aversion toward critical statements. These results highlight the risks and equity concerns posed to human authors and academic research if LLMs are deployed in the peer review cycle without adequate caution. On the other hand, revisions guided by LLM reviews yield quality gains in both LLM-based and human evaluations, illustrating the potential of the LLMs-as-reviewers for early-stage researchers and enhancing low-quality papers.

cs.CL↗

arXiv2025

Beyond Single Frames: Can LMMs Comprehend Temporal and Contextual Narratives in Image Sequences?

Large Multimodal Models (LMMs) have achieved remarkable success across various visual-language tasks. However, existing benchmarks predominantly focus on single-image understanding, leaving the analysis of image sequences largely unexplored. To address this limitation, we introduce StripCipher, a comprehensive benchmark designed to evaluate capabilities of LMMs to comprehend and reason over sequential images. StripCipher comprises a human-annotated dataset and three challenging subtasks: visual narrative comprehension, contextual frame prediction, and temporal narrative reordering. Our evaluation of 16 state-of-the-art LMMs, including GPT-4o and Qwen2.5VL, reveals a significant performance gap compared to human capabilities, particularly in tasks that require reordering shuffled sequential images. For instance, GPT-4o achieves only 23.93% accuracy in the reordering subtask, which is 56.07% lower than human performance. Further quantitative analysis discuss several factors, such as input format of images, affecting the performance of LLMs in sequential understanding, underscoring the fundamental challenges that remain in the development of LMMs.

cs.CL↗

arXiv2025

A Probabilistic Inference Scaling Theory for LLM Self-Correction

Large Language Models (LLMs) have demonstrated the capability to refine their generated answers through self-correction, enabling continuous performance improvement over multiple rounds. However, the mechanisms underlying how and why accuracy evolves during this iterative process remain unexplored. To fill this gap, we propose a probabilistic theory to model the dynamics of accuracy change and explain the performance improvements observed in multi-round self-correction. Through mathematical derivation, we establish that the accuracy after the $t^{th}$ round of self-correction is given by: $Acc_t = Upp - α^t(Upp - Acc_0),$ where $Acc_0$ denotes the initial accuracy, $Upp$ represents the upper bound of accuracy convergence, and $α$ determines the rate of convergence. Based on our theory, these parameters can be calculated and the predicted accuracy curve then can be obtained through only a single round of self-correction. Extensive experiments across diverse models and datasets demonstrate that our theoretical predictions align closely with empirical accuracy curves, validating the effectiveness of the theory. Our work provides a theoretical foundation for understanding LLM self-correction, thus paving the way for further explorations.

cs.CL↗

arXiv2025

How Far are LLMs from Being Our Digital Twins? A Benchmark for Persona-Based Behavior Chain Simulation

Recently, LLMs have garnered increasing attention across academic disciplines for their potential as human digital twins, virtual proxies designed to replicate individuals and autonomously perform tasks such as decision-making, problem-solving, and reasoning on their behalf. However, current evaluations of LLMs primarily emphasize dialogue simulation while overlooking human behavior simulation, which is crucial for digital twins. To address this gap, we introduce BehaviorChain, the first benchmark for evaluating LLMs' ability to simulate continuous human behavior. BehaviorChain comprises diverse, high-quality, persona-based behavior chains, totaling 15,846 distinct behaviors across 1,001 unique personas, each with detailed history and profile metadata. For evaluation, we integrate persona metadata into LLMs and employ them to iteratively infer contextually appropriate behaviors within dynamic scenarios provided by BehaviorChain. Comprehensive evaluation results demonstrated that even state-of-the-art models struggle with accurately simulating continuous human behavior.

cs.CL↗

arXiv2025

Towards Harmonized Uncertainty Estimation for Large Language Models

To facilitate robust and trustworthy deployment of large language models (LLMs), it is essential to quantify the reliability of their generations through uncertainty estimation. While recent efforts have made significant advancements by leveraging the internal logic and linguistic features of LLMs to estimate uncertainty scores, our empirical analysis highlights the pitfalls of these methods to strike a harmonized estimation between indication, balance, and calibration, which hinders their broader capability for accurate uncertainty estimation. To address this challenge, we propose CUE (Corrector for Uncertainty Estimation): A straightforward yet effective method that employs a lightweight model trained on data aligned with the target LLM's performance to adjust uncertainty scores. Comprehensive experiments across diverse models and tasks demonstrate its effectiveness, which achieves consistent improvements of up to 60% over existing methods.

cs.CL↗

arXiv2025

Reinforcement Pre-Training

In this work, we introduce Reinforcement Pre-Training (RPT) as a new scaling paradigm for large language models and reinforcement learning (RL). Specifically, we reframe next-token prediction as a reasoning task trained using RL, where it receives verifiable rewards for correctly predicting the next token for a given context. RPT offers a scalable method to leverage vast amounts of text data for general-purpose RL, rather than relying on domain-specific annotated answers. By incentivizing the capability of next-token reasoning, RPT significantly improves the language modeling accuracy of predicting the next tokens. Moreover, RPT provides a strong pre-trained foundation for further reinforcement fine-tuning. The scaling curves show that increased training compute consistently improves the next-token prediction accuracy. The results position RPT as an effective and promising scaling paradigm to advance language model pre-training.

cs.CL↗

arXiv2025

Large Language Models Struggle with Unreasonability in Math Problems

Large Language Models (LLMs) have shown remarkable success on a wide range of math and reasoning benchmarks. However, we observe that they often struggle when faced with unreasonable math problems. Instead of recognizing these issues, models frequently proceed as if the problem is well-posed, producing incorrect answers or falling into overthinking and verbose self-correction. To systematically investigate this overlooked vulnerability, we propose the \textbf{Unreasonable Math Problems (UMP)} benchmark, designed to evaluate LLMs' ability to detect and respond to unreasonable math problem statements. Based on extensive experiments covering 19 LLMs, we find that even state-of-the-art general models like GPT-4o achieve only a score of 0.6 on UMP. While reasoning models such as DeepSeek-R1 demonstrate a higher sensitivity to unreasonable inputs, this often comes at the cost of generating overly long and meaningless responses that fail to converge. We further explore prompting and fine-tuning methods, which offer partial improvements but also introduce trade-offs, shedding light on both the potential and limitations of LLMs in this challenging setting.

cs.CL↗

arXiv2025

From Mathematical Reasoning to Code: Generalization of Process Reward Models in Test-Time Scaling

Recent advancements in improving the reasoning capabilities of Large Language Models have underscored the efficacy of Process Reward Models (PRMs) in addressing intermediate errors through structured feedback mechanisms. This study analyzes PRMs from multiple perspectives, including training methodologies, scalability, and generalization capabilities. We investigate the interplay between pre-training and reward model training FLOPs to assess their influence on PRM efficiency and accuracy in complex reasoning tasks. Our analysis reveals a pattern of diminishing returns in performance with increasing PRM scale, highlighting the importance of balancing model size and computational cost. Furthermore, the diversity of training datasets significantly impacts PRM performance, emphasizing the importance of diverse data to enhance both accuracy and efficiency. We further examine test-time scaling strategies, identifying Monte Carlo Tree Search as the most effective method when computational resources are abundant, while Best-of-N Sampling serves as a practical alternative under resource-limited conditions. Notably, our findings indicate that PRMs trained on mathematical datasets exhibit performance comparable to those tailored for code generation, suggesting robust cross-domain generalization. Employing a gradient-based metric, we observe that PRMs exhibit a preference for selecting responses with similar underlying patterns, further informing their optimization.

cs.CL↗

arXiv2025

RICo: Refined In-Context Contribution for Automatic Instruction-Tuning Data Selection

Data selection for instruction tuning is crucial for improving the performance of large language models (LLMs) while reducing training costs. In this paper, we propose Refined Contribution Measurement with In-Context Learning (RICo), a novel gradient-free method that quantifies the fine-grained contribution of individual samples to both task-level and global-level model performance. RICo enables more accurate identification of high-contribution data, leading to better instruction tuning. We further introduce a lightweight selection paradigm trained on RICo scores, enabling scalable data selection with a strictly linear inference complexity. Extensive experiments on three LLMs across 12 benchmarks and 5 pairwise evaluation sets demonstrate the effectiveness of RICo. Remarkably, on LLaMA3.1-8B, models trained on 15% of RICo-selected data outperform full datasets by 5.42% points and exceed the best performance of widely used selection methods by 2.06% points. We further analyze high-contribution samples selected by RICo, which show both diverse tasks and appropriate difficulty levels, rather than just the hardest ones.

cs.CL↗

arXiv2025

SCoRE: Benchmarking Long-Chain Reasoning in Commonsense Scenarios

Currently, long-chain reasoning remains a key challenge for large language models (LLMs) because natural texts lack sufficient explicit reasoning data. However, existing benchmarks suffer from limitations such as narrow coverage, short reasoning paths, or high construction costs. We introduce SCoRE (Scenario-based Commonsense Reasoning Evaluation), a benchmark that synthesizes multi-hop questions from scenario schemas of entities, relations, and logical rules to assess long-chain commonsense reasoning. SCoRE contains 100k bilingual (Chinese-English) multiple-choice questions whose reasoning chains span 2-11 hops and are grouped into various difficulty levels. Each question is accompanied by fine-grained knowledge labels, explicit reasoning chains, and difficulty levels for diagnostic evaluation. Evaluation results on cutting-edge LLMs such as o3-mini and Deepseek R1 shows that even the best model attains only 69.78% accuracy on SCoRE (even only 47.91% on the hard set), with errors often stemming from rare knowledge, logical inconsistency, and over-interpretation of simple questions. SCoRE offers a scalable, extensible framework for evaluating and diagnosing the long-chain commonsense reasoning abilities of LLMs and guiding future advances in model design and training.

cs.CL↗

arXiv2025

Chain-of-Thought Tokens are Computer Program Variables

Chain-of-thoughts (CoT) requires large language models (LLMs) to generate intermediate steps before reaching the final answer, and has been proven effective to help LLMs solve complex reasoning tasks. However, the inner mechanism of CoT still remains largely unclear. In this paper, we empirically study the role of CoT tokens in LLMs on two compositional tasks: multi-digit multiplication and dynamic programming. While CoT is essential for solving these problems, we find that preserving only tokens that store intermediate results would achieve comparable performance. Furthermore, we observe that storing intermediate results in an alternative latent form will not affect model performance. We also randomly intervene some values in CoT, and notice that subsequent CoT tokens and the final answer would change correspondingly. These findings suggest that CoT tokens may function like variables in computer programs but with potential drawbacks like unintended shortcuts and computational complexity limits between tokens. The code and data are available at https://github.com/solitaryzero/CoTs_are_Variables.

cs.CL↗

arXiv2025

AdaMMS: Model Merging for Heterogeneous Multimodal Large Language Models with Unsupervised Coefficient Optimization

Recently, model merging methods have demonstrated powerful strengths in combining abilities on various tasks from multiple Large Language Models (LLMs). While previous model merging methods mainly focus on merging homogeneous models with identical architecture, they meet challenges when dealing with Multimodal Large Language Models (MLLMs) with inherent heterogeneous property, including differences in model architecture and the asymmetry in the parameter space. In this work, we propose AdaMMS, a novel model merging method tailored for heterogeneous MLLMs. Our method tackles the challenges in three steps: mapping, merging and searching. Specifically, we first design mapping function between models to apply model merging on MLLMs with different architecture. Then we apply linear interpolation on model weights to actively adapt the asymmetry in the heterogeneous MLLMs. Finally in the hyper-parameter searching step, we propose an unsupervised hyper-parameter selection method for model merging. As the first model merging method capable of merging heterogeneous MLLMs without labeled data, extensive experiments on various model combinations demonstrated that AdaMMS outperforms previous model merging methods on various vision-language benchmarks.

cs.CL↗

arXiv2025

Be a Multitude to Itself: A Prompt Evolution Framework for Red Teaming

Large Language Models (LLMs) have gained increasing attention for their remarkable capacity, alongside concerns about safety arising from their potential to produce harmful content. Red teaming aims to find prompts that could elicit harmful responses from LLMs, and is essential to discover and mitigate safety risks before real-world deployment. However, manual red teaming is both time-consuming and expensive, rendering it unscalable. In this paper, we propose RTPE, a scalable evolution framework to evolve red teaming prompts across both breadth and depth dimensions, facilitating the automatic generation of numerous high-quality and diverse red teaming prompts. Specifically, in-breadth evolving employs a novel enhanced in-context learning method to create a multitude of quality prompts, whereas in-depth evolving applies customized transformation operations to enhance both content and form of prompts, thereby increasing diversity. Extensive experiments demonstrate that RTPE surpasses existing representative automatic red teaming methods on both attack success rate and diversity. In addition, based on 4,800 red teaming prompts created by RTPE, we further provide a systematic analysis of 8 representative LLMs across 8 sensitive topics.

cs.CL↗

arXiv2024

Plug-and-Play Training Framework for Preference Optimization

Recently, preference optimization methods such as DPO have significantly enhanced large language models (LLMs) in wide tasks including dialogue and question-answering. However, current methods fail to account for the varying difficulty levels of training samples during preference optimization, leading to mediocre performance in tasks with high accuracy requirements, particularly in mathematical reasoning. To address this limitation, we propose a novel training framework, which employs multiple sampling to analyze output distributions, assign different weights to samples, and incorporate these weights into the preference optimization process. This plug-and-play approach enables LLMs to prioritize challenging examples during training, improving learning efficiency. Experimental results demonstrate that our framework integrates seamlessly with various preference optimization methods and achieves consistent improvements in mathematical reasoning tasks.

cs.CL↗

arXiv2024

Confidence v.s. Critique: A Decomposition of Self-Correction Capability for LLMs

Large Language Models (LLMs) can correct their self-generated responses, but a decline in accuracy after self-correction is also witnessed. To have a deeper understanding of self-correction, we endeavor to decompose, evaluate, and analyze the self-correction behaviors of LLMs. By enumerating and analyzing answer correctness before and after self-correction, we decompose the self-correction capability into confidence (being confident to correct answers) and critique (turning wrong answers to correct) capabilities, and propose two metrics from a probabilistic perspective to measure these 2 capabilities, along with another metric for overall self-correction capability evaluation. Based on our decomposition and evaluation metrics, we conduct extensive experiments and draw some empirical conclusions. For example, we find different models can exhibit distinct behaviors: some models are confident while others are more critical. We also find the trade-off between the two capabilities (i.e. improving one can lead to a decline in the other) when manipulating model self-correction behavior by prompts or in-context learning. Further, we find a simple yet efficient strategy to improve self-correction capability by transforming Supervision Fine-Tuning (SFT) data format, and our strategy outperforms vanilla SFT in both capabilities and achieves much higher accuracy after self-correction. Our code will be publicly available on GitHub.

cs.CL↗

arXiv2024

Language Models Encode the Value of Numbers Linearly

Large language models (LLMs) have exhibited impressive competence in various tasks, but their internal mechanisms on mathematical problems are still under-explored. In this paper, we study a fundamental question: how language models encode the value of numbers, a basic element in math. To study the question, we construct a synthetic dataset comprising addition problems and utilize linear probes to read out input numbers from the hidden states. Experimental results support the existence of encoded number values in LLMs on different layers, and these values can be extracted via linear probes. Further experiments show that LLMs store their calculation results in a similar manner, and we can intervene the output via simple vector additions, proving the causal connection between encoded numbers and language model outputs. Our research provides evidence that LLMs encode the value of numbers linearly, offering insights for better exploring, designing, and utilizing numeric information in LLMs.

cs.CL↗