Ning Wu

ByteDance Seed, Jiaze Chen, Tiantian Fan et al. · bytedance

We introduce Seed1.5-Thinking, capable of reasoning through thinking before responding, resulting in improved performance on a wide range of benchmarks. Seed1.5-Thinking achieves 86.7 on AIME 2024, 55.0 on Codeforces and 77.3 on GPQA, demonstrating excellent reasoning abilities in STEM and coding. Beyond reasoning tasks, the method demonstrates notable generalization across diverse domains. For instance, it surpasses DeepSeek R1 by 8% in win rate on non-reasoning tasks, indicating its broader applicability. Compared to other state-of-the-art reasoning models, Seed1.5-Thinking is a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters. As part of our effort to assess generalized reasoning, we develop two internal benchmarks, BeyondAIME and Codeforces, both of which will be publicly released to support future research. Model trial link: https://www.volcengine.com/experience/ark.

Nuo Chen, Zinan Zheng, Ning Wu et al.

Existing research predominantly focuses on developing powerful language learning models (LLMs) for mathematical reasoning within monolingual languages, with few explorations in preserving efficacy in a multilingual context. To bridge this gap, this paper pioneers exploring and training powerful Multilingual Math Reasoning (xMR) LLMs. Firstly, by utilizing translation, we construct the first multilingual math reasoning instruction dataset, MGSM8KInstruct, encompassing ten distinct languages, thus addressing the issue of training data scarcity in xMR tasks. Based on the collected dataset, we propose different training strategies to build powerful xMR LLMs, named MathOctopus, notably outperform conventional open-source LLMs and exhibit superiority over ChatGPT in few-shot scenarios. Notably, MathOctopus-13B reaches 47.6% accuracy which exceeds ChatGPT 46.3% on MGSM testset. Beyond remarkable results, we unearth several pivotal observations and insights from extensive experiments: (1) When extending the rejection sampling strategy to the multilingual context, it proves effective for model performances, albeit limited. (2) Employing parallel corpora for math Supervised Fine-Tuning (SFT) across multiple languages not only significantly enhances model performance multilingually but also elevates their monolingual performance. This indicates that crafting multilingual corpora can be regarded as a vital strategy for enhancing model performance in a specific language, especially in mathematical reasoning tasks. For instance, MathOctopus-7B improves its counterparts that trained on English from 42.2% to 50.8% on GSM8K testset. Codes are available at https://github.com/microsoft/MathOctopus.

Ning Wu, Ming Gong, Linjun Shou et al.

Text summarization has a wide range of applications in many scenarios. The evaluation of the quality of the generated text is a complex problem. A big challenge to language evaluation is that there is a clear divergence between existing metrics and human evaluation. A document summary's quality can be assessed by human annotators on various criteria, both objective ones like grammar and correctness, and subjective ones like informativeness, succinctness, and appeal. Most of the automatic evaluation methods like BLUE/ROUGE may be not able to adequately capture the above dimensions. In this paper, we propose a new evaluation framework based on LLMs, which provides a comprehensive evaluation framework by comparing generated text and reference text from both objective and subjective aspects. First, we propose to model objective and subjective dimensions of generated text based on roleplayers prompting mechanism. Furthermore, we introduce a context-based prompting mechanism that is able to generate dynamic roleplayer profiles based on input context. Finally, we design a multi-roleplayer prompting technology based on batch prompting and integrate multiple outputs into the final evaluation results. Experimental results on three real datasets for summarization show that our model is highly competitive and has a very high consistency with human annotators.

Tao Yang, Chuang Liu, Xiaofeng Ma et al.

Complex continuous or mixed joint distributions (e.g., P(Y | z_1, z_2, ..., z_N)) generally lack closed-form solutions, often necessitating approximations such as MCMC. This paper proposes Indeterminate Probability Theory (IPT), which makes the following contributions: (1) An observer-centered framework in which experimental outcomes are represented as distributions combining ground truth with observation error; (2) The introduction of three independence candidate axioms that enable a two-phase probabilistic inference framework; (3) The derivation of closed-form solutions for arbitrary complex joint distributions under this framework. Both the Indeterminate Probability Neural Network (IPNN) model and the non-neural multivariate time series forecasting application demonstrate IPT's effectiveness in modeling high-dimensional distributions, with successful validation up to 1000 dimensions. Importantly, IPT is consistent with classical probability theory and subsumes the frequentist equation in the limit of vanishing observation error.

Ning Wu, Rui Liu, Xinkun Lin et al.

Distilling demonstration effects into hidden-space interventions offers a lightweight alternative to full finetuning. However, existing multimodal variants are mostly evaluated on short-form tasks, where outputs end after a few tokens. Extending these methods to long-form generation exposes a fundamental yet underexamined limitation: token-level distillation implicitly treats all output tokens as equally informative, but long-form outputs are dominated by high-frequency template and grammatical tokens, while the tokens that actually determine output quality are sparsely distributed. In medical report generation (MRG), two such decisive tokens stand out: pathology-related tokens that determine diagnostic content, and the end-of-sequence (EOS) event that determines termination. Both receive insufficient supervision under uniform cross-entropy, and autoregressive decoding further compounds the problem by drifting away from teacher-forced trajectories. We propose DIVE, a frozen-backbone distillation framework that addresses long-form report generation through two complementary mechanisms matched to these failures. Decisive-token supervision restores supervision balance by upweighting the cross-entropy contribution of pathology-related tokens and the EOS event, ensuring that content fidelity and termination are learned during training rather than imposed at decoding time. State-conditioned dynamic steering replaces fixed open-loop residuals with hidden-state-dependent adapters, allowing the injected signal to adapt as decoding drifts. Experiments on MIMIC-CXR and CheXpert Plus with two medical VLM backbones show that DIVE consistently ranks among the strongest methods across lexical and clinical-proxy metrics. Our method achieves the best BLEU-4, ROUGE-L, and RadGraph F1 in all dataset--backbone settings, while remaining competitive on coarse label-level CheXbert F1.

Nuo Chen, Ning Wu, Jianhui Chang et al.

Utilizing large language models (LLMs) for data augmentation has yielded encouraging results in mathematical reasoning. However, these approaches face constraints in problem diversity, potentially restricting them to in-domain/distribution data generation. To this end, we propose ControlMath, an iterative method involving an equation-generator module and two LLM-based agents. The module creates diverse equations, which the Problem-Crafter agent then transforms into math word problems. The Reverse-Agent filters and selects high-quality data, adhering to the "less is more" principle, achieving better results with fewer data points. This approach enables the generation of diverse math problems, not limited to specific domains or distributions. As a result, we collect ControlMathQA, which involves 190k math word problems. Extensive results prove that combining our dataset with in-domain datasets like GSM8K can help improve the model's mathematical ability to generalize, leading to improved performances both within and beyond specific domains.

Nuo Chen, Ning Wu, Shining Liang et al.

This paper presents an in-depth analysis of Large Language Models (LLMs), focusing on LLaMA, a prominent open-source foundational model in natural language processing. Instead of assessing LLaMA through its generative output, we design multiple-choice tasks to probe its intrinsic understanding in high-order tasks such as reasoning and computation. We examine the model horizontally, comparing different sizes, and vertically, assessing different layers. We unveil several key and uncommon findings based on the designed probing tasks: (1) Horizontally, enlarging model sizes almost could not automatically impart additional knowledge or computational prowess. Instead, it can enhance reasoning abilities, especially in math problem solving, and helps reduce hallucinations, but only beyond certain size thresholds; (2) In vertical analysis, the lower layers of LLaMA lack substantial arithmetic and factual knowledge, showcasing logical thinking, multilingual and recognitive abilities, with top layers housing most computational power and real-world knowledge.

Shiping Yang, Jie Wu, Wenbiao Ding et al.

Robustness has become a critical attribute for the deployment of RAG systems in real-world applications. Existing research focuses on robustness to explicit noise (e.g., document semantics) but overlooks spurious features (a.k.a. implicit noise). While previous works have explored spurious features in LLMs, they are limited to specific features (e.g., formats) and narrow scenarios (e.g., ICL). In this work, we statistically confirm the presence of spurious features in the RAG paradigm, a robustness problem caused by the sensitivity of LLMs to semantic-agnostic features. Moreover, we provide a comprehensive taxonomy of spurious features and empirically quantify their impact through controlled experiments. Further analysis reveals that not all spurious features are harmful and they can even be beneficial sometimes. Extensive evaluation results across multiple LLMs suggest that spurious features are a widespread and challenging problem in the field of RAG. The code and dataset will be released to facilitate future research. We release all codes and data at: $\\\href{https://github.com/maybenotime/RAG-SpuriousFeatures}{https://github.com/maybenotime/RAG-SpuriousFeatures}$.

Lin Sun, Chuang Liu, Xiaofeng Ma et al.

Recent advancements in Large Language Models (LLMs) have demonstrated that Process Reward Models (PRMs) play a crucial role in enhancing model performance. However, training PRMs typically requires step-level labels, either manually annotated or automatically generated, which can be costly and difficult to obtain at scale. To address this challenge, we introduce FreePRM, a weakly supervised framework for training PRMs without access to ground-truth step-level labels. FreePRM first generates pseudo step-level labels based on the correctness of final outcome, and then employs Buffer Probability to eliminate impact of noise inherent in pseudo labeling. Experimental results show that FreePRM achieves an average F1 score of 53.0% on ProcessBench, outperforming fully supervised PRM trained on Math-Shepherd by +24.1%. Compared to other open-source PRMs, FreePRM outperforms upon RLHFlow-PRM-Mistral-8B (28.4%) by +24.6%, EurusPRM (31.3%) by +21.7%, and Skywork-PRM-7B (42.1%) by +10.9%. This work introduces a new paradigm in PRM training, significantly reducing reliance on costly step-level annotations while maintaining strong performance.

Haiping Yang, Huaxing Liu, Wei Wu et al.

Unmanned aerial vehicles (UAVs) are increasingly employed in diverse applications such as land surveying, material transport, and environmental monitoring. Following missions like data collection or inspection, UAVs must land safely at docking stations for storage or recharging, which is an essential requirement for ensuring operational continuity. However, accurate landing remains challenging due to factors like GPS signal interference. To address this issue, we propose a deviation warning system for UAV landings, powered by a novel vision-based model called AeroLite-MDNet. This model integrates a multiscale fusion module for robust cross-scale object detection and incorporates a segmentation branch for efficient orientation estimation. We introduce a new evaluation metric, Average Warning Delay (AWD), to quantify the system's sensitivity to landing deviations. Furthermore, we contribute a new dataset, UAVLandData, which captures real-world landing deviation scenarios to support training and evaluation. Experimental results show that our system achieves an AWD of 0.7 seconds with a deviation detection accuracy of 98.6\%, demonstrating its effectiveness in enhancing UAV landing reliability. Code will be available at https://github.com/ITTTTTI/Maskyolo.git

Jie Deng, Hanshuang Tong, Jun Li et al.

Large language models (LLMs) have made impressive strides in mathematical reasoning, often fine-tuned using rejection sampling that retains only correct reasoning trajectories. While effective, this paradigm treats supervision as a binary filter that systematically excludes teacher-generated errors, leaving a gap in how reasoning failures are modeled during training. In this paper, we propose TrajFusion, a fine-tuning strategy that reframes rejection sampling as a structured supervision construction process. Specifically, TrajFusion forms fused trajectories that explicitly model trial-and-error reasoning by interleaving selected incorrect trajectories with reflection prompts and correct trajectories. The length of each fused sample is adaptively controlled based on the frequency and diversity of teacher errors, providing richer supervision for challenging problems while safely reducing to vanilla rejection sampling fine-tuning (RFT) when error signals are uninformative. TrajFusion requires no changes to the architecture or training objective. Extensive experiments across multiple math benchmarks demonstrate that TrajFusion consistently outperforms RFT, particularly on challenging and long-form reasoning problems.

Hanshuang Tong, Jun Li, Ning Wu et al.

Recent advancements in large language models (LLMs) have opened new pathways for many domains. However, the full potential of LLMs in financial investments remains largely untapped. There are two main challenges for typical deep learning-based methods for quantitative finance. First, they struggle to fuse textual and numerical information flexibly for stock movement prediction. Second, traditional methods lack clarity and interpretability, which impedes their application in scenarios where the justification for predictions is essential. To solve the above challenges, we propose Ploutos, a novel financial LLM framework that consists of PloutosGen and PloutosGPT. The PloutosGen contains multiple primary experts that can analyze different modal data, such as text and numbers, and provide quantitative strategies from different perspectives. Then PloutosGPT combines their insights and predictions and generates interpretable rationales. To generate accurate and faithful rationales, the training strategy of PloutosGPT leverage rearview-mirror prompting mechanism to guide GPT-4 to generate rationales, and a dynamic token weighting mechanism to finetune LLM by increasing key tokens weight. Extensive experiments show our framework outperforms the state-of-the-art methods on both prediction accuracy and interpretability.

Ning Wu, Han Huang, Yueting Xu et al.

Architecture performance evaluation is the most time-consuming part of neural architecture search (NAS). Zero-Shot NAS accelerates the evaluation by utilizing zero-cost proxies instead of training. Though effective, existing zero-cost proxies require invoking backpropagations or running networks on input data, making it difficult to further accelerate the computation of proxies. To alleviate this issue, architecture topologies are used to evaluate the performance of networks in this study. We prove that particular architectural topologies decrease the local entropy of feature maps, which degrades specific features to a bias, thereby reducing network performance. Based on this proof, architectural topologies are utilized to quantify the suppression of local entropy decrease (SED) as a data-free and running-free proxy. Experimental results show that SED outperforms most state-of-the-art proxies in terms of architecture selection on five benchmarks, with computation time reduced by three orders of magnitude. We further compare the SED-based NAS with state-of-the-art proxies. SED-based NAS selects the architecture with higher accuracy and fewer parameters in only one second. The theoretical analyses of local entropy and experimental results demonstrate that the suppression of local entropy decrease facilitates selecting optimal architectures in Zero-Shot NAS.

Weihao Liu, Ning Wu, Shiping Yang et al.

Large Language Models (LLMs) frequently show distracted attention due to irrelevant information in the input, which severely impairs their long-context capabilities. Inspired by recent studies on the effectiveness of retrieval heads in long-context factutality, we aim at addressing this distraction issue through improving such retrieval heads directly. We propose Multi-Document Attention Focusing (MuDAF), a novel method that explicitly optimizes the attention distribution at the head level through contrastive learning. According to the experimental results, MuDAF can significantly improve the long-context question answering performance of LLMs, especially in multi-document question answering. Extensive evaluations on retrieval scores and attention visualizations show that MuDAF possesses great potential in making attention heads more focused on relevant information and reducing attention distractions.

Lin Sun, Chuang Liu, Peng Liu et al.

Direct Preference Optimization (DPO) have emerged as a popular method for aligning Large Language Models (LLMs) with human preferences. While DPO effectively preserves the relative ordering between chosen and rejected responses through pairwise ranking losses, it often neglects absolute reward magnitudes. This oversight can decrease the likelihood of chosen responses and increase the risk of generating out-of-distribution responses, leading to poor performance. We term this issue Degraded Chosen Responses (DCR).To address this issue, we propose Balanced Preference Optimization (BPO), a novel framework that dynamically balances the optimization of chosen and rejected responses through two key components: balanced reward margin and gap adaptor. Unlike previous methods, BPO can fundamentally resolve DPO's DCR issue, without introducing additional constraints to the loss function. Experimental results on multiple mathematical reasoning tasks show that BPO significantly outperforms DPO, improving accuracy by +10.1% with Llama-3.1-8B-Instruct (18.8% to 28.9%) and +11.7% with Qwen2.5-Math-7B (35.0% to 46.7%). It also surpasses DPO variants by +3.6% over IPO (43.1%), +5.0% over SLiC (41.7%), and +3.1% over Cal-DPO (43.6%) on the same model. Remarkably, our algorithm requires only a single line of code modification, making it simple to implement and fully compatible with existing DPO-based frameworks.

Weihao Liu, Ning Wu, Wenbiao Ding et al.

Truthfulness stands out as an essential challenge for Large Language Models (LLMs). Although many works have developed various ways for truthfulness enhancement, they seldom focus on truthfulness in multilingual scenarios. Meanwhile, contemporary multilingual aligning technologies struggle to balance numerous languages and often exhibit serious truthfulness gaps across different languages, especially those that differ greatly from English. In our work, we extend truthfulness evaluation to multilingual contexts and propose a practical method for cross-lingual truthfulness transfer called Fact-aware Multilingual Selective Synergy (FaMSS). FaMSS is able to select an optimal subset of all tested languages by language bias and transfer contributions, and then employ translation instruction tuning for cross-lingual truthfulness transfer. Experimental results demonstrate that our approach can effectively reduce the multilingual representation disparity and boost cross-lingual truthfulness transfer of LLMs.

Yaobo Liang, Nan Duan, Yeyun Gong et al.

In this paper, we introduce XGLUE, a new benchmark dataset that can be used to train large-scale cross-lingual pre-trained models using multilingual and bilingual corpora and evaluate their performance across a diverse set of cross-lingual tasks. Comparing to GLUE(Wang et al., 2019), which is labeled in English for natural language understanding tasks only, XGLUE has two main advantages: (1) it provides 11 diversified tasks that cover both natural language understanding and generation scenarios; (2) for each task, it provides labeled data in multiple languages. We extend a recent cross-lingual pre-trained model Unicoder(Huang et al., 2019) to cover both understanding and generation tasks, which is evaluated on XGLUE as a strong baseline. We also evaluate the base versions (12-layer) of Multilingual BERT, XLM and XLM-R for comparison.

Jingyuan Wang, Ning Wu, Wayne Xin Zhao et al.

Personalized Route Recommendation (PRR) aims to generate user-specific route suggestions in response to users' route queries. Early studies cast the PRR task as a pathfinding problem on graphs, and adopt adapted search algorithms by integrating heuristic strategies. Although these methods are effective to some extent, they require setting the cost functions with heuristics. In addition, it is difficult to utilize useful context information in the search procedure. To address these issues, we propose using neural networks to automatically learn the cost functions of a classic heuristic algorithm, namely A* algorithm, for the PRR task. Our model consists of two components. First, we employ attention-based Recurrent Neural Networks (RNN) to model the cost from the source to the candidate location by incorporating useful context information. Instead of learning a single cost value, the RNN component is able to learn a time-varying vectorized representation for the moving state of a user. Second, we propose to use a value network for estimating the cost from a candidate location to the destination. For capturing structural characteristics, the value network is built on top of improved graph attention networks by incorporating the moving state of a user and other context information. The two components are integrated in a principled way for deriving a more accurate cost of a candidate location. Extensive experiment results on three real-world datasets have shown the effectiveness and robustness of the proposed model.