Hui Liu

Fali Wang, Jihai Chen, Shuhua Yang et al.

Test-Time Scaling (TTS) improves large language models (LLMs) by allocating additional computation during inference, typically through parallel, sequential, or hybrid scaling. However, prior studies often assume fixed collaboration architectures (e.g., topologies) and single-model usage, overlooking that optimal architectures and model combinations can vary across tasks. Therefore, we study the novel problem of searching for compute-optimal model combinations and architectures in TTS under a fixed budget. We formalize it as a multi-LLM collaboration graph, where nodes encode roles and LLM model assignments, and edges capture information flow. This problem is challenging because (i) the combinatorial search space is prohibitively large, and (ii) task-specific requirements demand tailored designs. To address these, we reformulate the problem as probabilistic graph optimization and, through pilot experiments, derive three empirical insights into TTS collaboration graphs. Guided by these insights, we propose Agent-REINFORCE, an LLM-agent-augmented framework that mirrors the REINFORCE pipeline by mapping sampling-gradient-update to sampling-feedback-update, where feedback serves as a textual gradient to update the probabilistic graph and efficiently search for optimal multi-LLM collaboration graphs. Experiments show that Agent-REINFORCE outperforms both traditional and LLM-based baselines in sample efficiency and search performance, and effectively identifies optimal graphs under joint objectives of accuracy and inference latency.

Yueqing Xi, Yifan Bai, Huasen Luo et al.

As artificial intelligence permeates judicial forensics, ensuring the veracity and traceability of legal question answering (QA) has become critical. Conventional large language models (LLMs) are prone to hallucination, risking misleading guidance in legal consultation, while static knowledge bases struggle to keep pace with frequently updated statutes and case law. We present a hybrid legal QA agent tailored for judicial settings that integrates retrieval-augmented generation (RAG) with multi-model ensembling to deliver reliable, auditable, and continuously updatable counsel. The system prioritizes retrieval over generation: when a trusted legal repository yields relevant evidence, answers are produced via RAG; otherwise, multiple LLMs generate candidates that are scored by a specialized selector, with the top-ranked answer returned. High-quality outputs then undergo human review before being written back to the repository, enabling dynamic knowledge evolution and provenance tracking. Experiments on the Law\_QA dataset show that our hybrid approach significantly outperforms both a single-model baseline and a vanilla RAG pipeline on F1, ROUGE-L, and an LLM-as-a-Judge metric. Ablations confirm the complementary contributions of retrieval prioritization, model ensembling, and the human-in-the-loop update mechanism. The proposed system demonstrably reduces hallucination while improving answer quality and legal compliance, advancing the practical landing of media forensics technologies in judicial scenarios.

Yadong Li, Jun Liu, Tao Zhang et al.

We introduce Baichuan-Omni-1.5, an omni-modal model that not only has omni-modal understanding capabilities but also provides end-to-end audio generation capabilities. To achieve fluent and high-quality interaction across modalities without compromising the capabilities of any modality, we prioritized optimizing three key aspects. First, we establish a comprehensive data cleaning and synthesis pipeline for multimodal data, obtaining about 500B high-quality data (text, audio, and vision). Second, an audio-tokenizer (Baichuan-Audio-Tokenizer) has been designed to capture both semantic and acoustic information from audio, enabling seamless integration and enhanced compatibility with MLLM. Lastly, we designed a multi-stage training strategy that progressively integrates multimodal alignment and multitask fine-tuning, ensuring effective synergy across all modalities. Baichuan-Omni-1.5 leads contemporary models (including GPT4o-mini and MiniCPM-o 2.6) in terms of comprehensive omni-modal capabilities. Notably, it achieves results comparable to leading models such as Qwen2-VL-72B across various multimodal medical benchmarks.

Xiaoke Huang, Juncheng Wu, Hui Liu et al.

Test-time scaling has emerged as a powerful technique for enhancing the reasoning capabilities of large language models. However, its effectiveness in medical reasoning remains uncertain, as the medical domain fundamentally differs from mathematical tasks in terms of knowledge representation and decision-making processes. In this paper, we provide the first comprehensive investigation of test-time scaling for medical reasoning and present m1, a simple yet effective approach that increases a model's medical reasoning capability at inference. Our evaluation across diverse medical tasks demonstrates that test-time scaling consistently enhances medical reasoning, enabling lightweight fine-tuned models under 10B parameters to establish new state-of-the-art performance, while our 32B model rivals previous 70B-scale medical LLMs. However, we identify an optimal reasoning token budget of approximately 4K, beyond which performance may degrade due to overthinking. Budget forcing, which extends test-time computation through iterative prompts, helps models double-check answers but does not necessarily improve the overall medical QA performance and, in some cases, even introduces errors into previously correct responses. Our case-by-case analysis identifies insufficient medical knowledge as a key bottleneck that prevents further performance gains through test-time scaling. We find that increasing data scale, improving data quality, and expanding model capacity consistently enhance medical knowledge grounding, enabling continued performance improvements, particularly on challenging medical benchmarks where smaller models reach saturation. These findings underscore fundamental differences between medical and mathematical reasoning in LLMs, highlighting that enriched medical knowledge, other than increased reasoning depth alone, is essential for realizing the benefits of test-time scaling.

Fali Wang, Hui Liu, Zhenwei Dai et al.

Test-time scaling (TTS) enhances the performance of large language models (LLMs) by allocating additional compute resources during inference. However, existing research primarily investigates TTS in single-stage tasks; while many real-world problems are multi-stage complex tasks, composed of a sequence of heterogeneous subtasks with each subtask requires LLM of specific capability. Therefore, we study a novel problem: the test-time compute-optimal scaling in multi-stage complex tasks, aiming to select suitable models and allocate budgets per subtask to maximize overall performance. TTS in multi-stage tasks introduces two fundamental challenges: (i) The combinatorial search space of model and budget allocations, combined with the high cost of inference, makes brute-force search impractical. (ii) The optimal model and budget allocations across subtasks are interdependent, increasing the complexity of the compute-optimal search. To address this gap, we conduct extensive pilot experiments on four tasks across six datasets, deriving three empirical insights characterizing the behavior of LLMs in multi-stage complex tasks. Informed by these insights, we propose AgentTTS, an LLM-agent-based framework that autonomously searches for compute-optimal allocations through iterative feedback-driven interactions with the execution environment. Experimental results demonstrate that AgentTTS significantly outperforms traditional and other LLM-based baselines in search efficiency, and shows improved robustness to varying training set sizes and enhanced interpretability.

Jiankun Zhang, Shenglai Zeng, Jie Ren et al.

Multimodal Retrieval-Augmented Generation (MRAG) systems enhance LMMs by integrating external multimodal databases, but introduce unexplored privacy vulnerabilities. While text-based RAG privacy risks have been studied, multimodal data presents unique challenges. We provide the first systematic analysis of MRAG privacy vulnerabilities across vision-language and speech-language modalities. Using a novel compositional structured prompt attack in a black-box setting, we demonstrate how attackers can extract private information by manipulating queries. Our experiments reveal that LMMs can both directly generate outputs resembling retrieved content and produce descriptions that indirectly expose sensitive information, highlighting the urgent need for robust privacy-preserving MRAG techniques.

Yanjie Jiang, Hui Liu, Nan Niu et al.

High-quality and large-scale repositories of real bugs and their concise patches collected from real-world applications are critical for research in software engineering community. In such a repository, each real bug is explicitly associated with its fix. Therefore, on one side, the real bugs and their fixes} may inspire novel approaches for finding, locating, and repairing software bugs; on the other side, the real bugs and their fixes are indispensable for rigorous and meaningful evaluation of approaches for software testing, fault localization, and program repair. To this end, a number of such repositories, e.g., Defects4J, have been proposed. However, such repositories are rather small because their construction involves expensive human intervention. Although bug-fixing code commits as well as associated test cases could be retrieved from version control systems automatically, existing approaches could not yet automatically extract concise bug-fixing patches from bug-fixing commits because such commits often involve bug-irrelevant changes. In this paper, we propose an automatic approach, called BugBuilder, to extracting complete and concise bug-fixing patches from human-written patches in version control systems. It excludes refactorings by detecting refactorings involved in bug-fixing commits, and reapplying detected refactorings on the faulty version. It enumerates all subsets of the remaining part and validates them on test cases. If none of the subsets has the potential to be a complete bug-fixing patch, the remaining part as a whole is taken as a complete and concise bug-fixing patch. Evaluation results on 809 real bug-fixing commits in Defects4J suggest that BugBuilder successfully generated complete and concise bug-fixing patches for forty percent of the bug-fixing commits, and its precision (99%) was even higher than human experts.

Xiangyu Zhao, Haochen Liu, Hui Liu et al.

Practical large-scale recommender systems usually contain thousands of feature fields from users, items, contextual information, and their interactions. Most of them empirically allocate a unified dimension to all feature fields, which is memory inefficient. Thus it is highly desired to assign different embedding dimensions to different feature fields according to their importance and predictability. Due to the large amounts of feature fields and the nuanced relationship between embedding dimensions with feature distributions and neural network architectures, manually allocating embedding dimensions in practical recommender systems can be very difficult. To this end, we propose an AutoML based framework (AutoDim) in this paper, which can automatically select dimensions for different feature fields in a data-driven fashion. Specifically, we first proposed an end-to-end differentiable framework that can calculate the weights over various dimensions for feature fields in a soft and continuous manner with an AutoML based optimization algorithm; then we derive a hard and discrete embedding component architecture according to the maximal weights and retrain the whole recommender framework. We conduct extensive experiments on benchmark datasets to validate the effectiveness of the AutoDim framework.