Zhiwei He

Wenxiang Jiao, Jen-tse Huang, Wenxuan Wang et al. · pku, tencent-ai

Large language models (LLMs) like ChatGPT have exhibited remarkable abilities on a wide range of natural language processing~(NLP) tasks, including various machine translation abilities accomplished during chat. However, these models are only accessible through restricted APIs, which creates barriers to new research and advancements in the field. Therefore, we propose ParroT, a framework to enhance and regulate the translation abilities during chat based on open-source LLMs (e.g., LLaMA), human-written translation and feedback data. Specifically, ParroT reformulates translation data into the instruction-following style, and introduces a "$\mathbf{Hint}$" field for incorporating extra requirements to regulate the translation process. Accordingly, we propose three instruction types for finetuning ParroT models, including translation instruction, contrastive instruction, and error-guided instruction. Experiments on Flores subsets and WMT22 test sets suggest that translation instruction improves the translation performance of vanilla LLMs significantly while error-guided instruction can lead to further improvement, which demonstrates the importance of learning from low-quality translations annotated by humans. We also demonstrate the potential of automatic evaluation tools in providing quality information of translations, when constructing error-guided instructions for directions that lack human annotation data. Please refer to our Github project for more implementation details: https://github.com/wxjiao/ParroT

Zhuosheng Zhang, Yao Yao, Aston Zhang et al.

Large language models (LLMs) have dramatically enhanced the field of language intelligence, as demonstrably evidenced by their formidable empirical performance across a spectrum of complex reasoning tasks. Additionally, theoretical proofs have illuminated their emergent reasoning capabilities, providing a compelling showcase of their advanced cognitive abilities in linguistic contexts. Critical to their remarkable efficacy in handling complex reasoning tasks, LLMs leverage the intriguing chain-of-thought (CoT) reasoning techniques, obliging them to formulate intermediate steps en route to deriving an answer. The CoT reasoning approach has not only exhibited proficiency in amplifying reasoning performance but also in enhancing interpretability, controllability, and flexibility. In light of these merits, recent research endeavors have extended CoT reasoning methodologies to nurture the development of autonomous language agents, which adeptly adhere to language instructions and execute actions within varied environments. This survey paper orchestrates a thorough discourse, penetrating vital research dimensions, encompassing: (i) the foundational mechanics of CoT techniques, with a focus on elucidating the circumstances and justification behind its efficacy; (ii) the paradigm shift in CoT; and (iii) the burgeoning of language agents fortified by CoT approaches. Prospective research avenues envelop explorations into generalization, efficiency, customization, scaling, and safety. This paper caters to a wide audience, including beginners seeking comprehensive knowledge of CoT reasoning and language agents, as well as experienced researchers interested in foundational mechanics and engaging in cutting-edge discussions on these topics. A repository for the related papers is available at https://github.com/Zoeyyao27/CoT-Igniting-Agent.

Jiahao Nie, Zhiwei He, Yuxiang Yang et al.

Current 3D single object tracking methods are typically based on VoteNet, a 3D region proposal network. Despite the success, using a single seed point feature as the cue for offset learning in VoteNet prevents high-quality 3D proposals from being generated. Moreover, seed points with different importance are treated equally in the voting process, aggravating this defect. To address these issues, we propose a novel global-local transformer voting scheme to provide more informative cues and guide the model pay more attention on potential seed points, promoting the generation of high-quality 3D proposals. Technically, a global-local transformer (GLT) module is employed to integrate object- and patch-aware prior into seed point features to effectively form strong feature representation for geometric positions of the seed points, thus providing more robust and accurate cues for offset learning. Subsequently, a simple yet effective training strategy is designed to train the GLT module. We develop an importance prediction branch to learn the potential importance of the seed points and treat the output weights vector as a training constraint term. By incorporating the above components together, we exhibit a superior tracking method GLT-T. Extensive experiments on challenging KITTI and NuScenes benchmarks demonstrate that GLT-T achieves state-of-the-art performance in the 3D single object tracking task. Besides, further ablation studies show the advantages of the proposed global-local transformer voting scheme over the original VoteNet. Code and models will be available at https://github.com/haooozi/GLT-T.

Jiahao Nie, Zhiwei He, Yuxiang Yang et al.

Two-stage point-to-box network acts as a critical role in the recent popular 3D Siamese tracking paradigm, which first generates proposals and then predicts corresponding proposal-wise scores. However, such a network suffers from tedious hyper-parameter tuning and task misalignment, limiting the tracking performance. Towards these concerns, we propose a simple yet effective one-stage point-to-box network for point cloud-based 3D single object tracking. It synchronizes 3D proposal generation and center-ness score prediction by a parallel predictor without tedious hyper-parameters. To guide a task-aligned score ranking of proposals, a center-aware focal loss is proposed to supervise the training of the center-ness branch, which enhances the network's discriminative ability to distinguish proposals of different quality. Besides, we design a binary target classifier to identify target-relevant points. By integrating the derived classification scores with the center-ness scores, the resulting network can effectively suppress interference proposals and further mitigate task misalignment. Finally, we present a novel one-stage Siamese tracker OSP2B equipped with the designed network. Extensive experiments on challenging benchmarks including KITTI and Waymo SOT Dataset show that our OSP2B achieves leading performance with a considerable real-time speed.Code will be available at https://github.com/haooozi/OSP2B.

Zhiwei He, Xing Wang, Rui Wang et al. · tencent-ai

Back-translation is a critical component of Unsupervised Neural Machine Translation (UNMT), which generates pseudo parallel data from target monolingual data. A UNMT model is trained on the pseudo parallel data with translated source, and translates natural source sentences in inference. The source discrepancy between training and inference hinders the translation performance of UNMT models. By carefully designing experiments, we identify two representative characteristics of the data gap in source: (1) style gap (i.e., translated vs. natural text style) that leads to poor generalization capability; (2) content gap that induces the model to produce hallucination content biased towards the target language. To narrow the data gap, we propose an online self-training approach, which simultaneously uses the pseudo parallel data {natural source, translated target} to mimic the inference scenario. Experimental results on several widely-used language pairs show that our approach outperforms two strong baselines (XLM and MASS) by remedying the style and content gaps.

Jiahao Nie, Fei Xie, Sifan Zhou et al.

3D single object tracking (SOT) methods based on appearance matching has long suffered from insufficient appearance information incurred by incomplete, textureless and semantically deficient LiDAR point clouds. While motion paradigm exploits motion cues instead of appearance matching for tracking, it incurs complex multi-stage processing and segmentation module. In this paper, we first provide in-depth explorations on motion paradigm, which proves that (\textbf{i}) it is feasible to directly infer target relative motion from point clouds across consecutive frames; (\textbf{ii}) fine-grained information comparison between consecutive point clouds facilitates target motion modeling. We thereby propose to perform part-to-part motion modeling for consecutive point clouds and introduce a novel tracking framework, termed \textbf{P2P}. The novel framework fuses each corresponding part information between consecutive point clouds, effectively exploring detailed information changes and thus modeling accurate target-related motion cues. Following this framework, we present P2P-point and P2P-voxel models, incorporating implicit and explicit part-to-part motion modeling by point- and voxel-based representation, respectively. Without bells and whistles, P2P-voxel sets a new state-of-the-art performance ($\sim$\textbf{89\%}, \textbf{72\%} and \textbf{63\%} precision on KITTI, NuScenes and Waymo Open Dataset, respectively). Moreover, under the same point-based representation, P2P-point outperforms the previous motion tracker M$^2$Track by \textbf{3.3\%} and \textbf{6.7\%} on the KITTI and NuScenes, while running at a considerably high speed of \textbf{107 Fps} on a single RTX3090 GPU. The source code and pre-trained models are available at https://github.com/haooozi/P2P.

Zhiwei He, Xing Wang, Zhaopeng Tu et al. · tencent-ai

This paper describes Tencent AI Lab - Shanghai Jiao Tong University (TAL-SJTU) Low-Resource Translation systems for the WMT22 shared task. We participate in the general translation task on English$\Leftrightarrow$Livonian. Our system is based on M2M100 with novel techniques that adapt it to the target language pair. (1) Cross-model word embedding alignment: inspired by cross-lingual word embedding alignment, we successfully transfer a pre-trained word embedding to M2M100, enabling it to support Livonian. (2) Gradual adaptation strategy: we exploit Estonian and Latvian as auxiliary languages for many-to-many translation training and then adapt to English-Livonian. (3) Data augmentation: to enlarge the parallel data for English-Livonian, we construct pseudo-parallel data with Estonian and Latvian as pivot languages. (4) Fine-tuning: to make the most of all available data, we fine-tune the model with the validation set and online back-translation, further boosting the performance. In model evaluation: (1) We find that previous work underestimated the translation performance of Livonian due to inconsistent Unicode normalization, which may cause a discrepancy of up to 14.9 BLEU score. (2) In addition to the standard validation set, we also employ round-trip BLEU to evaluate the models, which we find more appropriate for this task. Finally, our unconstrained system achieves BLEU scores of 17.0 and 30.4 for English to/from Livonian.

Jiahao Nie, Zhiwei He, Yuxiang Yang et al.

Siamese trackers based on 3D region proposal network (RPN) have shown remarkable success with deep Hough voting. However, using a single seed point feature as the cue for voting fails to produce high-quality 3D proposals. Additionally, the equal treatment of seed points in the voting process, regardless of their significance, exacerbates this limitation. To address these challenges, we propose a novel transformer-based voting scheme to generate better proposals. Specifically, a global-local transformer (GLT) module is devised to integrate object- and patch-aware geometric priors into seed point features, resulting in robust and accurate cues for offset learning of seed points. To train the GLT module, we introduce an importance prediction branch that learns the potential importance weights of seed points as a training constraint. Incorporating this transformer-based voting scheme into 3D RPN, a novel Siamese method dubbed GLT-T is developed for 3D single object tracking on point clouds. Moreover, we identify that the highest-scored proposal in the Siamese paradigm may not be the most accurate proposal, which limits tracking performance. Towards this concern, we approach the binary score prediction task as a ranking problem, and design a target-aware ranking loss and a localization-aware ranking loss to produce accurate ranking of proposals. With the ranking losses, we further present GLT-T++, an enhanced version of GLT-T. Extensive experiments on multiple benchmarks demonstrate that our GLT-T and GLT-T++ outperform state-of-the-art methods in terms of tracking accuracy while maintaining a real-time inference speed. The source code will be made available at https://github.com/haooozi/GLT-T.

Tian Liang, Zhiwei He, Jen-tse Huang et al. · pku, tencent-ai

The automatic evaluation of LLM-based agent intelligence is critical in developing advanced LLM-based agents. Although considerable effort has been devoted to developing human-annotated evaluation datasets, such as AlpacaEval, existing techniques are costly, time-consuming, and lack adaptability. In this paper, inspired by the popular language game ``Who is Spy'', we propose to use the word guessing game to assess the intelligence performance of LLMs. Given a word, the LLM is asked to describe the word and determine its identity (spy or not) based on its and other players' descriptions. Ideally, an advanced agent should possess the ability to accurately describe a given word using an aggressive description while concurrently maximizing confusion in the conservative description, enhancing its participation in the game. To this end, we first develop DEEP to evaluate LLMs' expression and disguising abilities. DEEP requires LLM to describe a word in aggressive and conservative modes. We then introduce SpyGame, an interactive multi-agent framework designed to assess LLMs' intelligence through participation in a competitive language-based board game. Incorporating multi-agent interaction, SpyGame requires the target LLM to possess linguistic skills and strategic thinking, providing a more comprehensive evaluation of LLMs' human-like cognitive abilities and adaptability in complex communication situations. The proposed evaluation framework is very easy to implement. We collected words from multiple sources, domains, and languages and used the proposed evaluation framework to conduct experiments. Extensive experiments demonstrate that the proposed DEEP and SpyGame effectively evaluate the capabilities of various LLMs, capturing their ability to adapt to novel situations and engage in strategic communication.

Jiahao Nie, Han Wu, Zhiwei He et al.

Siamese tracking paradigm has achieved great success, providing effective appearance discrimination and size estimation by the classification and regression. While such a paradigm typically optimizes the classification and regression independently, leading to task misalignment (accurate prediction boxes have no high target confidence scores). In this paper, to alleviate this misalignment, we propose a novel tracking paradigm, called SiamLA. Within this paradigm, a series of simple, yet effective localization-aware components are introduced, to generate localization-aware target confidence scores. Specifically, with the proposed localization-aware dynamic label (LADL) loss and localization-aware label smoothing (LALS) strategy, collaborative optimization between the classification and regression is achieved, enabling classification scores to be aware of location state, not just appearance similarity. Besides, we propose a separate localization branch, centered on a localization-aware feature aggregation (LAFA) module, to produce location quality scores to further modify the classification scores. Consequently, the resulting target confidence scores, are more discriminative for the location state, allowing accurate prediction boxes tend to be predicted as high scores. Extensive experiments are conducted on six challenging benchmarks, including GOT-10k, TrackingNet, LaSOT, TNL2K, OTB100 and VOT2018. Our SiamLA achieves state-of-the-art performance in terms of both accuracy and efficiency. Furthermore, a stability analysis reveals that our tracking paradigm is relatively stable, implying the paradigm is potential to real-world applications.

MiniMax, Aili Chen, Aonian Li et al.

We introduce the MiniMax-M2 series, a family of Mixture-of-Experts language models built around the principle that mini activations can unleash maximum real-world intelligence. The flagship M2 contains 229.9B total parameters with only 9.8B activated per token. Designed end-to-end for agentic deployment, the M2 series rests on three components: (i) agent-driven data pipelines producing large-scale, verifiable trajectories across agentic coding and agentic cowork, each grounded in an executable workspace and an artifact-aligned reward; (ii) Forge, a scalable agent-native RL system that adapts to long-horizon agent trajectories, paired with windowed-FIFO scheduling, prefix-tree merging, inference optimization, and a clean training-inference-agent decoupling that supports both white-box and black-box agents; (iii) the latest M2.7 checkpoint takes an early step toward self-evolution -- autonomously debugging training runs and modifying its own scaffold. Across M2 through M2.7, this combination translates a mini-activation footprint into frontier-tier performance on agentic coding, deep search, office-task, and reasoning benchmarks.

Yue Wang, Qiuzhi Liu, Jiahao Xu et al.

Large language models (LLMs) such as OpenAI's o1 have demonstrated remarkable abilities in complex reasoning tasks by scaling test-time compute and exhibiting human-like deep thinking. However, we identify a phenomenon we term underthinking, where o1-like LLMs frequently switch between different reasoning thoughts without sufficiently exploring promising paths to reach a correct solution. This behavior leads to inadequate depth of reasoning and decreased performance, particularly on challenging mathematical problems. To systematically analyze this issue, we conduct experiments on three challenging test sets and two representative open-source o1-like models, revealing that frequent thought switching correlates with incorrect responses. We introduce a novel metric to quantify underthinking by measuring token efficiency in incorrect answers. To address underthinking, we propose a decoding strategy with thought switching penalty TIP that discourages premature transitions between thoughts, encouraging deeper exploration of each reasoning path. Experimental results demonstrate that our approach improves accuracy across challenging datasets without requiring model fine-tuning. Our findings contribute to understanding reasoning inefficiencies in o1-like LLMs and offer a practical solution to enhance their problem-solving capabilities.

Zhiwei He, Tian Liang, Jiahao Xu et al.

Reinforcement learning (RL) with large language models shows promise in complex reasoning. However, its progress is hindered by the lack of large-scale training data that is sufficiently challenging, contamination-free and verifiable. To this end, we introduce DeepMath-103K, a large-scale mathematical dataset designed with high difficulty (primarily levels 5-9), rigorous decontamination against numerous benchmarks, and verifiable answers for rule-based RL reward. It further includes three distinct R1 solutions adaptable for diverse training paradigms such as supervised fine-tuning (SFT). Spanning a wide range of mathematical topics, DeepMath-103K fosters the development of generalizable and advancing reasoning. Notably, models trained on DeepMath-103K achieve state-of-the-art results on challenging mathematical benchmarks and demonstrate generalization beyond math such as biology, physics and chemistry, underscoring its broad efficacy. Data: https://huggingface.co/datasets/zwhe99/DeepMath-103K.

Yuxuan Lu, Jiahao Nie, Zhiwei He et al.

Current LiDAR point cloud-based 3D single object tracking (SOT) methods typically rely on point-based representation network. Despite demonstrated success, such networks suffer from some fundamental problems: 1) It contains pooling operation to cope with inherently disordered point clouds, hindering the capture of 3D spatial information that is useful for tracking, a regression task. 2) The adopted set abstraction operation hardly handles density-inconsistent point clouds, also preventing 3D spatial information from being modeled. To solve these problems, we introduce a novel tracking framework, termed VoxelTrack. By voxelizing inherently disordered point clouds into 3D voxels and extracting their features via sparse convolution blocks, VoxelTrack effectively models precise and robust 3D spatial information, thereby guiding accurate position prediction for tracked objects. Moreover, VoxelTrack incorporates a dual-stream encoder with cross-iterative feature fusion module to further explore fine-grained 3D spatial information for tracking. Benefiting from accurate 3D spatial information being modeled, our VoxelTrack simplifies tracking pipeline with a single regression loss. Extensive experiments are conducted on three widely-adopted datasets including KITTI, NuScenes and Waymo Open Dataset. The experimental results confirm that VoxelTrack achieves state-of-the-art performance (88.3%, 71.4% and 63.6% mean precision on the three datasets, respectively), and outperforms the existing trackers with a real-time speed of 36 Fps on a single TITAN RTX GPU. The source code and model will be released.

Leyi Pan, Aiwei Liu, Zhiwei He et al. · berkeley, tsinghua

LLM watermarking, which embeds imperceptible yet algorithmically detectable signals in model outputs to identify LLM-generated text, has become crucial in mitigating the potential misuse of large language models. However, the abundance of LLM watermarking algorithms, their intricate mechanisms, and the complex evaluation procedures and perspectives pose challenges for researchers and the community to easily experiment with, understand, and assess the latest advancements. To address these issues, we introduce MarkLLM, an open-source toolkit for LLM watermarking. MarkLLM offers a unified and extensible framework for implementing LLM watermarking algorithms, while providing user-friendly interfaces to ensure ease of access. Furthermore, it enhances understanding by supporting automatic visualization of the underlying mechanisms of these algorithms. For evaluation, MarkLLM offers a comprehensive suite of 12 tools spanning three perspectives, along with two types of automated evaluation pipelines. Through MarkLLM, we aim to support researchers while improving the comprehension and involvement of the general public in LLM watermarking technology, fostering consensus and driving further advancements in research and application. Our code is available at https://github.com/THU-BPM/MarkLLM.

Wenhong Zhu, Hongkun Hao, Zhiwei He et al.

We are currently in an era of fierce competition among various large language models (LLMs) continuously pushing the boundaries of benchmark performance. However, genuinely assessing the capabilities of these LLMs has become a challenging and critical issue due to potential data contamination, and it wastes dozens of time and effort for researchers and engineers to download and try those contaminated models. To save our precious time, we propose a novel and useful method, Clean-Eval, which mitigates the issue of data contamination and evaluates the LLMs in a cleaner manner. Clean-Eval employs an LLM to paraphrase and back-translate the contaminated data into a candidate set, generating expressions with the same meaning but in different surface forms. A semantic detector is then used to filter the generated low-quality samples to narrow down this candidate set. The best candidate is finally selected from this set based on the BLEURT score. According to human assessment, this best candidate is semantically similar to the original contamination data but expressed differently. All candidates can form a new benchmark to evaluate the model. Our experiments illustrate that Clean-Eval substantially restores the actual evaluation results on contaminated LLMs under both few-shot learning and fine-tuning scenarios.

Tian Liang, Wenxiang Jiao, Zhiwei He et al.

Large Reasoning Models (LRMs) have demonstrated impressive capabilities but suffer from cognitive inefficiencies like ``overthinking'' simple problems and ``underthinking'' complex ones. While existing methods that use supervised fine-tuning~(SFT) or reinforcement learning~(RL) with token-length rewards can improve efficiency, they often do so at the cost of accuracy. This paper introduces \textbf{DeepCompress}, a novel framework that simultaneously enhances both the accuracy and efficiency of LRMs. We challenge the prevailing approach of consistently favoring shorter reasoning paths, showing that longer responses can contain a broader range of correct solutions for difficult problems. DeepCompress employs an adaptive length reward mechanism that dynamically classifies problems as ``Simple'' or ``Hard'' in real-time based on the model's evolving capability. It encourages shorter, more efficient reasoning for ``Simple'' problems while promoting longer, more exploratory thought chains for ``Hard'' problems. This dual-reward strategy enables the model to autonomously adjust its Chain-of-Thought (CoT) length, compressing reasoning for well-mastered problems and extending it for those it finds challenging. Experimental results on challenging mathematical benchmarks show that DeepCompress consistently outperforms baseline methods, achieving superior accuracy while significantly improving token efficiency.

Zhiwei He, Xing Wang, Wenxiang Jiao et al. · tencent-ai

Insufficient modeling of human preferences within the reward model is a major obstacle for leveraging human feedback to improve translation quality. Fortunately, quality estimation (QE), which predicts the quality of a given translation without reference, has achieved impressive alignment with human evaluations in the last two years. In this work, we investigate the potential of employing the QE model as the reward model to predict human preferences for feedback training. We first identify the overoptimization problem during QE-based feedback training, manifested as an increase in reward while translation quality declines. We examine the problem and argue that the vulnerability of the QE model might lead to high rewards for incorrect translations, resulting in overoptimization and error propagation. To address the problem, we adopt a simple yet effective method that uses heuristic rules to detect the incorrect translations and assigns a penalty term to the reward scores of them. Experimental results show that the proposed QE-based feedback training achieves consistent and significant improvements across various settings, further verified through human preference studies. Our subsequent analysis demonstrates the high data efficiency of the proposed QE-based feedback training: it outperforms systems using larger parallel corpora by a small amount of monolingual data. Our code is available at: https://github.com/zwhe99/FeedbackMT

Zhiwei He, Binglin Zhou, Hongkun Hao et al. · tencent-ai, tsinghua

Text watermarking technology aims to tag and identify content produced by large language models (LLMs) to prevent misuse. In this study, we introduce the concept of cross-lingual consistency in text watermarking, which assesses the ability of text watermarks to maintain their effectiveness after being translated into other languages. Preliminary empirical results from two LLMs and three watermarking methods reveal that current text watermarking technologies lack consistency when texts are translated into various languages. Based on this observation, we propose a Cross-lingual Watermark Removal Attack (CWRA) to bypass watermarking by first obtaining a response from an LLM in a pivot language, which is then translated into the target language. CWRA can effectively remove watermarks, decreasing the AUCs to a random-guessing level without performance loss. Furthermore, we analyze two key factors that contribute to the cross-lingual consistency in text watermarking and propose X-SIR as a defense method against CWRA. Code: https://github.com/zwhe99/X-SIR.

Xiaolin Xing, Zhiwei He, Haoyu Xu et al.

This paper investigates the cross-lingual inconsistencies observed in Large Language Models (LLMs), such as ChatGPT, Llama, and Baichuan, which have shown exceptional performance in various Natural Language Processing (NLP) tasks. Despite their successes, these models often exhibit significant inconsistencies when processing the same concepts across different languages. This study focuses on three primary questions: the existence of cross-lingual inconsistencies in LLMs, the specific aspects in which these inconsistencies manifest, and the correlation between cross-lingual consistency and multilingual capabilities of LLMs.To address these questions, we propose an innovative evaluation method for Cross-lingual Semantic Consistency (xSC) using the LaBSE model. We further introduce metrics for Cross-lingual Accuracy Consistency (xAC) and Cross-lingual Timeliness Consistency (xTC) to comprehensively assess the models' performance regarding semantic, accuracy, and timeliness inconsistencies. By harmonizing these metrics, we provide a holistic measurement of LLMs' cross-lingual consistency. Our findings aim to enhance the understanding and improvement of multilingual capabilities and interpretability in LLMs, contributing to the development of more robust and reliable multilingual language models.

Yansi Li, Jiahao Xu, Tian Liang et al.

Enhancing the reasoning capabilities of large language models (LLMs), particularly for complex tasks requiring multi-step logical deductions, remains a significant challenge. Traditional inference time scaling methods utilize scalar reward signals from process reward models to evaluate candidate reasoning steps, but these scalar rewards lack the nuanced qualitative information essential for understanding and justifying each step. In this paper, we propose a novel inference-time scaling approach -- stepwise natural language self-critique (PANEL), which employs self-generated natural language critiques as feedback to guide the step-level search process. By generating rich, human-readable critiques for each candidate reasoning step, PANEL retains essential qualitative information, facilitating better-informed decision-making during inference. This approach bypasses the need for task-specific verifiers and the associated training overhead, making it broadly applicable across diverse tasks. Experimental results on challenging reasoning benchmarks, including AIME and GPQA, demonstrate that PANEL significantly enhances reasoning performance, outperforming traditional scalar reward-based methods. Our code is available at https://github.com/puddingyeah/PANEL to support and encourage future research in this promising field.

Zhiwei He, Zhaopeng Tu, Xing Wang et al.

Low-rank adaptation (LoRA) has been prominently employed for parameter-efficient fine-tuning of large language models (LLMs). However, the limited expressive capacity of LoRA, stemming from the low-rank constraint, has been recognized as a bottleneck, particularly in rigorous tasks like code generation and mathematical reasoning. To address this limitation, we introduce Rank-Sharing Low-Rank Adaptation (RaSA), an innovative extension that enhances the expressive capacity of LoRA by leveraging partial rank sharing across layers. By forming a shared rank pool and applying layer-specific weighting, RaSA effectively increases the number of ranks without augmenting parameter overhead. Our theoretically grounded and empirically validated approach demonstrates that RaSA not only maintains the core advantages of LoRA but also significantly boosts performance in challenging code and math tasks. Code, data and scripts are available at: https://github.com/zwhe99/RaSA.

Xingyu Chen, Jiahao Xu, Tian Liang et al.

The remarkable performance of models like the OpenAI o1 can be attributed to their ability to emulate human-like long-time thinking during inference. These models employ extended chain-of-thought (CoT) processes, exploring multiple strategies to enhance problem-solving capabilities. However, a critical question remains: How to intelligently and efficiently scale computational resources during testing. This paper presents the first comprehensive study on the prevalent issue of overthinking in these models, where excessive computational resources are allocated for simple problems with minimal benefit. We introduce novel efficiency metrics from both outcome and process perspectives to evaluate the rational use of computational resources by o1-like models. Using a self-training paradigm, we propose strategies to mitigate overthinking, streamlining reasoning processes without compromising accuracy. Experimental results show that our approach successfully reduces computational overhead while preserving model performance across a range of testsets with varying difficulty levels, such as GSM8K, MATH500, GPQA, and AIME.

Jiahao Nie, Zhiwei He, Xudong Lv et al.

Category-specific models are provenly valuable methods in 3D single object tracking (SOT) regardless of Siamese or motion-centric paradigms. However, such over-specialized model designs incur redundant parameters, thus limiting the broader applicability of 3D SOT task. This paper first introduces unified models that can simultaneously track objects across all categories using a single network with shared model parameters. Specifically, we propose to explicitly encode distinct attributes associated to different object categories, enabling the model to adapt to cross-category data. We find that the attribute variances of point cloud objects primarily occur from the varying size and shape (e.g., large and square vehicles v.s. small and slender humans). Based on this observation, we design a novel point set representation learning network inheriting transformer architecture, termed AdaFormer, which adaptively encodes the dynamically varying shape and size information from cross-category data in a unified manner. We further incorporate the size and shape prior derived from the known template targets into the model's inputs and learning objective, facilitating the learning of unified representation. Equipped with such designs, we construct two category-unified models SiamCUT and MoCUT.Extensive experiments demonstrate that SiamCUT and MoCUT exhibit strong generalization and training stability. Furthermore, our category-unified models outperform the category-specific counterparts by a significant margin (e.g., on KITTI dataset, 12% and 3% performance gains on the Siamese and motion paradigms). Our code will be available.

Tongxin Yuan, Zhiwei He, Lingzhong Dong et al.

Large language models (LLMs) have exhibited great potential in autonomously completing tasks across real-world applications. Despite this, these LLM agents introduce unexpected safety risks when operating in interactive environments. Instead of centering on the harmlessness of LLM-generated content in most prior studies, this work addresses the imperative need for benchmarking the behavioral safety of LLM agents within diverse environments. We introduce R-Judge, a benchmark crafted to evaluate the proficiency of LLMs in judging and identifying safety risks given agent interaction records. R-Judge comprises 569 records of multi-turn agent interaction, encompassing 27 key risk scenarios among 5 application categories and 10 risk types. It is of high-quality curation with annotated safety labels and risk descriptions. Evaluation of 11 LLMs on R-Judge shows considerable room for enhancing the risk awareness of LLMs: The best-performing model, GPT-4o, achieves 74.42% while no other models significantly exceed the random. Moreover, we reveal that risk awareness in open agent scenarios is a multi-dimensional capability involving knowledge and reasoning, thus challenging for LLMs. With further experiments, we find that fine-tuning on safety judgment significantly improve model performance while straightforward prompting mechanisms fail. R-Judge is publicly available at https://github.com/Lordog/R-Judge.

Tian Liang, Zhiwei He, Wenxiang Jiao et al.

Modern large language models (LLMs) like ChatGPT have shown remarkable performance on general language tasks but still struggle on complex reasoning tasks, which drives the research on cognitive behaviors of LLMs to explore human-like problem-solving strategies. Along this direction, one representative strategy is self-reflection, which asks an LLM to refine the solution with the feedback generated by itself iteratively. However, our study shows that such reflection-style methods suffer from the Degeneration-of-Thought (DoT) problem: once the LLM has established confidence in its solutions, it is unable to generate novel thoughts later through reflection even if its initial stance is incorrect. To address the DoT problem, we propose a Multi-Agent Debate (MAD) framework, in which multiple agents express their arguments in the state of "tit for tat" and a judge manages the debate process to obtain a final solution. Clearly, our MAD framework encourages divergent thinking in LLMs which would be helpful for tasks that require deep levels of contemplation. Experiment results on two challenging datasets, commonsense machine translation and counter-intuitive arithmetic reasoning, demonstrate the effectiveness of our MAD framework. Extensive analyses suggest that the adaptive break of debate and the modest level of "tit for tat" state are required for MAD to obtain good performance. Moreover, we find that LLMs might not be a fair judge if different LLMs are used for agents. Code is available at https://github.com/Skytliang/Multi-Agents-Debate.

Zhiwei He, Tian Liang, Wenxiang Jiao et al.

Large language models (LLMs) have demonstrated impressive capabilities in general scenarios, exhibiting a level of aptitude that approaches, in some aspects even surpasses, human-level intelligence. Among their numerous skills, the translation abilities of LLMs have received considerable attention. Compared to typical machine translation that focuses solely on source-to-target mapping, LLM-based translation can potentially mimic the human translation process which might take preparatory steps to ensure high-quality translation. This work explores this possibility by proposing the MAPS framework, which stands for Multi-Aspect Prompting and Selection. Specifically, we enable LLMs first to analyze the given source sentence and induce three aspects of translation-related knowledge: keywords, topics, and relevant demonstrations to guide the final translation process. Moreover, we employ a selection mechanism based on quality estimation to filter out noisy and unhelpful knowledge. Both automatic (3 LLMs x 11 directions x 2 automatic metrics) and human evaluation (preference study and MQM) demonstrate the effectiveness of MAPS. Further analysis shows that by mimicking the human translation process, MAPS reduces various translation errors such as hallucination, ambiguity, mistranslation, awkward style, untranslated text, and omission. Source code is available at https://github.com/zwhe99/MAPS-mt.

Tian Xia, Zhiwei He, Tong Ren et al.

Bargaining is an important and unique part of negotiation between humans. As LLM-driven agents learn to negotiate and act like real humans, how to evaluate agents' bargaining abilities remains an open problem. For the first time, we formally described the Bargaining task as an asymmetric incomplete information game, defining the gains of the Buyer and Seller in multiple bargaining processes. It allows us to quantitatively assess an agent's performance in the Bargain task. We collected a real product price dataset, AmazonHistoryPrice, and conducted evaluations of various LLM agents' bargaining abilities. We find that playing a Buyer is much harder than a Seller, and increasing model size can not effectively improve the Buyer's performance. To address the challenge, we propose a novel approach called OG-Narrator that integrates a deterministic Offer Generator to control the price range of Buyer's offers, and an LLM Narrator to create natural language sentences for generated offers. Experimental results show that OG-Narrator improves the buyer's deal rates from 26.67% to 88.88% and brings a ten times multiplication of profits on all baselines, even a model that has not been aligned.

Ke Ji, Jiahao Xu, Tian Liang et al.

Improving the reasoning capabilities of large language models (LLMs) typically requires supervised fine-tuning with labeled data or computationally expensive sampling. We introduce Unsupervised Prefix Fine-Tuning (UPFT), which leverages the observation of Prefix Self-Consistency -- the shared initial reasoning steps across diverse solution trajectories -- to enhance LLM reasoning efficiency. By training exclusively on the initial prefix substrings (as few as 8 tokens), UPFT removes the need for labeled data or exhaustive sampling. Experiments on reasoning benchmarks show that UPFT matches the performance of supervised methods such as Rejection Sampling Fine-Tuning, while reducing training time by 75% and sampling cost by 99%. Further analysis reveals that errors tend to appear in later stages of the reasoning process and that prefix-based training preserves the model's structural knowledge. This work demonstrates how minimal unsupervised fine-tuning can unlock substantial reasoning gains in LLMs, offering a scalable and resource-efficient alternative to conventional approaches.

Wenhong Zhu, Hongkun Hao, Zhiwei He et al.

Current language models decode text token by token according to probabilistic distribution, and determining the appropriate candidates for the next token is crucial to ensure generation quality. This study introduces adaptive decoding, a mechanism that dynamically empowers language models to ascertain a sensible candidate set during generation. Specifically, we introduce an entropy-based metric called confidence and conceptualize determining the optimal candidate set as a confidence-increasing process. The rationality of including a token in the candidate set is assessed by leveraging the increment of confidence. Experimental results reveal that our method balances diversity and coherence well. The human evaluation shows that our method can generate human-preferred text. Additionally, our method can potentially improve the reasoning ability of language models.

Xiaoyuan Liu, Tian Liang, Zhiwei He et al.

Large Language Models (LLMs) show great promise in complex reasoning, with Reinforcement Learning with Verifiable Rewards (RLVR) being a key enhancement strategy. However, a prevalent issue is ``superficial self-reflection'', where models fail to robustly verify their own outputs. We introduce RISE (Reinforcing Reasoning with Self-Verification), a novel online RL framework designed to tackle this. RISE explicitly and simultaneously trains an LLM to improve both its problem-solving and self-verification abilities within a single, integrated RL process. The core mechanism involves leveraging verifiable rewards from an outcome verifier to provide on-the-fly feedback for both solution generation and self-verification tasks. In each iteration, the model generates solutions, then critiques its own on-policy generated solutions, with both trajectories contributing to the policy update. Extensive experiments on diverse mathematical reasoning benchmarks show that RISE consistently improves model's problem-solving accuracy while concurrently fostering strong self-verification skills. Our analyses highlight the advantages of online verification and the benefits of increased verification compute. Additionally, RISE models exhibit more frequent and accurate self-verification behaviors during reasoning. These advantages reinforce RISE as a flexible and effective path towards developing more robust and self-aware reasoners.

Wenhong Zhu, Zhiwei He, Xiaofeng Wang et al.

Aligning language models (LMs) with human preferences has become a key area of research, enabling these models to meet diverse user needs better. Inspired by weak-to-strong generalization, where a strong LM fine-tuned on labels generated by a weaker model can consistently outperform its weak supervisor, we extend this idea to model alignment. In this work, we observe that the alignment behavior in weaker models can be effectively transferred to stronger models and even exhibit an amplification effect. Based on this insight, we propose a method called Weak-to-Strong Preference Optimization (WSPO), which achieves strong model alignment by learning the distribution differences before and after the alignment of the weak model. Experiments demonstrate that WSPO delivers outstanding performance, improving the win rate of Qwen2-7B-Instruct on Arena-Hard from 39.70 to 49.60, achieving a remarkable 47.04 length-controlled win rate on AlpacaEval 2, and scoring 7.33 on MT-bench. Our results suggest that using the weak model to elicit a strong model with a high alignment ability is feasible.

Ziyin Zhang, Jiahao Xu, Zhiwei He et al.

Theorem proving serves as a major testbed for evaluating complex reasoning abilities in large language models (LLMs). However, traditional automated theorem proving (ATP) approaches rely heavily on formal proof systems that poorly align with LLMs' strength derived from informal, natural language knowledge acquired during pre-training. In this work, we propose DeepTheorem, a comprehensive informal theorem-proving framework exploiting natural language to enhance LLM mathematical reasoning. DeepTheorem includes a large-scale benchmark dataset consisting of 121K high-quality IMO-level informal theorems and proofs spanning diverse mathematical domains, rigorously annotated for correctness, difficulty, and topic categories, accompanied by systematically constructed verifiable theorem variants. We devise a novel reinforcement learning strategy (RL-Zero) explicitly tailored to informal theorem proving, leveraging the verified theorem variants to incentivize robust mathematical inference. Additionally, we propose comprehensive outcome and process evaluation metrics examining proof correctness and the quality of reasoning steps. Extensive experimental analyses demonstrate DeepTheorem significantly improves LLM theorem-proving performance compared to existing datasets and supervised fine-tuning protocols, achieving state-of-the-art accuracy and reasoning quality. Our findings highlight DeepTheorem's potential to fundamentally advance automated informal theorem proving and mathematical exploration.

Peisong Wang, Ruotian Ma, Bang Zhang et al.

Large language models (LLMs) excel at logical and algorithmic reasoning, yet their emotional intelligence (EQ) still lags far behind their cognitive prowess. While reinforcement learning from verifiable rewards (RLVR) has advanced in other domains, its application to dialogue-especially for emotional intelligence-remains underexplored. In this work, we introduce RLVER, the first end-to-end reinforcement learning framework that leverages verifiable emotion rewards from simulated users to cultivate higher-order empathetic abilities in LLMs. Within this framework, self-consistent affective simulated users engage in dialogue rollouts and produce deterministic emotion scores during conversations, serving as reward signals to guide the LLM's learning. Fine-tuning publicly available Qwen2.5-7B-Instruct model with PPO boosts its Sentient-Benchmark score from 13.3 to 79.2 while largely preserving mathematical and coding competence. Extensive experiments reveal that: (i) RLVER consistently improves multiple dialogue capabilities; (ii) Thinking and non-thinking models show distinct trends--thinking models excel in empathy and insight, while non-thinking models favor action; (iii) GRPO often yields stable gains, while PPO can push certain capabilities to a higher ceiling; (iv) More challenging environments are not always better-moderate ones can yield stronger outcomes. Our results show that RLVER is a practical route toward emotionally intelligent and broadly capable language agents.

Yiming Ai, Zhiwei He, Ziyin Zhang et al.

In this study, we delve into the validity of conventional personality questionnaires in capturing the human-like personality traits of Large Language Models (LLMs). Our objective is to assess the congruence between the personality traits LLMs claim to possess and their demonstrated tendencies in real-world scenarios. By conducting an extensive examination of LLM outputs against observed human response patterns, we aim to understand the disjunction between self-knowledge and action in LLMs.

Mengru Wang, Xingyu Chen, Yue Wang et al.

Mixture-of-Experts (MoE) architectures within Large Reasoning Models (LRMs) have achieved impressive reasoning capabilities by selectively activating experts to facilitate structured cognitive processes. Despite notable advances, existing reasoning models often suffer from cognitive inefficiencies like overthinking and underthinking. To address these limitations, we introduce a novel inference-time steering methodology called Reinforcing Cognitive Experts (RICE), designed to improve reasoning performance without additional training or complex heuristics. Leveraging normalized Pointwise Mutual Information (nPMI), we systematically identify specialized experts, termed ''cognitive experts'' that orchestrate meta-level reasoning operations characterized by tokens like ''<think>''. Empirical evaluations with leading MoE-based LRMs (DeepSeek-R1 and Qwen3-235B) on rigorous quantitative and scientific reasoning benchmarks demonstrate noticeable and consistent improvements in reasoning accuracy, cognitive efficiency, and cross-domain generalization. Crucially, our lightweight approach substantially outperforms prevalent reasoning-steering techniques, such as prompt design and decoding constraints, while preserving the model's general instruction-following skills. These results highlight reinforcing cognitive experts as a promising, practical, and interpretable direction to enhance cognitive efficiency within advanced reasoning models.

Zhengsheng Guo, Zhiwei He, Wenxiang Jiao et al. · tencent-ai

Motivated by the success of unsupervised neural machine translation (UNMT), we introduce an unsupervised sign language translation and generation network (USLNet), which learns from abundant single-modality (text and video) data without parallel sign language data. USLNet comprises two main components: single-modality reconstruction modules (text and video) that rebuild the input from its noisy version in the same modality and cross-modality back-translation modules (text-video-text and video-text-video) that reconstruct the input from its noisy version in the different modality using back-translation procedure.Unlike the single-modality back-translation procedure in text-based UNMT, USLNet faces the cross-modality discrepancy in feature representation, in which the length and the feature dimension mismatch between text and video sequences. We propose a sliding window method to address the issues of aligning variable-length text with video sequences. To our knowledge, USLNet is the first unsupervised sign language translation and generation model capable of generating both natural language text and sign language video in a unified manner. Experimental results on the BBC-Oxford Sign Language dataset (BOBSL) and Open-Domain American Sign Language dataset (OpenASL) reveal that USLNet achieves competitive results compared to supervised baseline models, indicating its effectiveness in sign language translation and generation.

Ziyin Zhang, Jiahao Xu, Tian Liang et al.

Conventional speculative decoding (SD) methods utilize a predefined length policy for proposing drafts, which implies the premise that the target model smoothly accepts the proposed draft tokens. However, reality deviates from this assumption: the oracle draft length varies significantly, and the fixed-length policy hardly satisfies such a requirement. Moreover, such discrepancy is further exacerbated in scenarios involving complex reasoning and long-form generation, particularly under test-time scaling for reasoning-specialized models. Through both theoretical and empirical estimation, we establish that the discrepancy between the draft and target models can be approximated by the draft model's prediction entropy: a high entropy indicates a low acceptance rate of draft tokens, and vice versa. Based on this insight, we propose SVIP: Self-Verification Length Policy for Long-Context Speculative Decoding, which is a training-free dynamic length policy for speculative decoding systems that adaptively determines the lengths of draft sequences by referring to the draft entropy. Experimental results on mainstream SD benchmarks as well as reasoning-heavy benchmarks demonstrate the superior performance of SVIP, achieving up to 17% speedup on MT-Bench at 8K context compared with fixed draft lengths, and 22% speedup for QwQ in long-form reasoning.

Yiming Ai, Zhiwei He, Kai Yu et al.

Tense inconsistency frequently occurs in machine translation. However, there are few criteria to assess the model's mastery of tense prediction from a linguistic perspective. In this paper, we present a parallel tense test set, containing French-English 552 utterances. We also introduce a corresponding benchmark, tense prediction accuracy. With the tense test set and the benchmark, researchers are able to measure the tense consistency performance of machine translation systems for the first time.