Bolin Ding

Dawei Gao, Haibin Wang, Yaliang Li et al.

Large language models (LLMs) have emerged as a new paradigm for Text-to-SQL task. However, the absence of a systematical benchmark inhibits the development of designing effective, efficient and economic LLM-based Text-to-SQL solutions. To address this challenge, in this paper, we first conduct a systematical and extensive comparison over existing prompt engineering methods, including question representation, example selection and example organization, and with these experimental results, we elaborate their pros and cons. Based on these findings, we propose a new integrated solution, named DAIL-SQL, which refreshes the Spider leaderboard with 86.6% execution accuracy and sets a new bar. To explore the potential of open-source LLM, we investigate them in various scenarios, and further enhance their performance with supervised fine-tuning. Our explorations highlight open-source LLMs' potential in Text-to-SQL, as well as the advantages and disadvantages of the supervised fine-tuning. Additionally, towards an efficient and economic LLM-based Text-to-SQL solution, we emphasize the token efficiency in prompt engineering and compare the prior studies under this metric. We hope that our work provides a deeper understanding of Text-to-SQL with LLMs, and inspires further investigations and broad applications.

Yupeng Hou, Shanlei Mu, Wayne Xin Zhao et al.

In order to develop effective sequential recommenders, a series of sequence representation learning (SRL) methods are proposed to model historical user behaviors. Most existing SRL methods rely on explicit item IDs for developing the sequence models to better capture user preference. Though effective to some extent, these methods are difficult to be transferred to new recommendation scenarios, due to the limitation by explicitly modeling item IDs. To tackle this issue, we present a novel universal sequence representation learning approach, named UniSRec. The proposed approach utilizes the associated description text of items to learn transferable representations across different recommendation scenarios. For learning universal item representations, we design a lightweight item encoding architecture based on parametric whitening and mixture-of-experts enhanced adaptor. For learning universal sequence representations, we introduce two contrastive pre-training tasks by sampling multi-domain negatives. With the pre-trained universal sequence representation model, our approach can be effectively transferred to new recommendation domains or platforms in a parameter-efficient way, under either inductive or transductive settings. Extensive experiments conducted on real-world datasets demonstrate the effectiveness of the proposed approach. Especially, our approach also leads to a performance improvement in a cross-platform setting, showing the strong transferability of the proposed universal SRL method. The code and pre-trained model are available at: https://github.com/RUCAIBox/UniSRec.

Zhen Wang, Weirui Kuang, Ce Zhang et al. · eth-zurich

Hyperparameter optimization (HPO) is crucial for machine learning algorithms to achieve satisfactory performance, whose progress has been boosted by related benchmarks. Nonetheless, existing efforts in benchmarking all focus on HPO for traditional centralized learning while ignoring federated learning (FL), a promising paradigm for collaboratively learning models from dispersed data. In this paper, we first identify some uniqueness of HPO for FL algorithms from various aspects. Due to this uniqueness, existing HPO benchmarks no longer satisfy the need to compare HPO methods in the FL setting. To facilitate the research of HPO in the FL setting, we propose and implement a benchmark suite FedHPO-B that incorporates comprehensive FL tasks, enables efficient function evaluations, and eases continuing extensions. We also conduct extensive experiments based on FedHPO-B to benchmark a few HPO methods. We open-source FedHPO-B at https://github.com/alibaba/FederatedScope/tree/master/benchmark/FedHPOB.

Yuexiang Xie, Zhen Wang, Dawei Gao et al.

Although remarkable progress has been made by existing federated learning (FL) platforms to provide infrastructures for development, these platforms may not well tackle the challenges brought by various types of heterogeneity, including the heterogeneity in participants' local data, resources, behaviors and learning goals. To fill this gap, in this paper, we propose a novel FL platform, named FederatedScope, which employs an event-driven architecture to provide users with great flexibility to independently describe the behaviors of different participants. Such a design makes it easy for users to describe participants with various local training processes, learning goals and backends, and coordinate them into an FL course with synchronous or asynchronous training strategies. Towards an easy-to-use and flexible platform, FederatedScope enables rich types of plug-in operations and components for efficient further development, and we have implemented several important components to better help users with privacy protection, attack simulation and auto-tuning. We have released FederatedScope at https://github.com/alibaba/FederatedScope to promote academic research and industrial deployment of federated learning in a wide range of scenarios.

Zhen Wang, Weirui Kuang, Yuexiang Xie et al.

The incredible development of federated learning (FL) has benefited various tasks in the domains of computer vision and natural language processing, and the existing frameworks such as TFF and FATE has made the deployment easy in real-world applications. However, federated graph learning (FGL), even though graph data are prevalent, has not been well supported due to its unique characteristics and requirements. The lack of FGL-related framework increases the efforts for accomplishing reproducible research and deploying in real-world applications. Motivated by such strong demand, in this paper, we first discuss the challenges in creating an easy-to-use FGL package and accordingly present our implemented package FederatedScope-GNN (FS-G), which provides (1) a unified view for modularizing and expressing FGL algorithms; (2) comprehensive DataZoo and ModelZoo for out-of-the-box FGL capability; (3) an efficient model auto-tuning component; and (4) off-the-shelf privacy attack and defense abilities. We validate the effectiveness of FS-G by conducting extensive experiments, which simultaneously gains many valuable insights about FGL for the community. Moreover, we employ FS-G to serve the FGL application in real-world E-commerce scenarios, where the attained improvements indicate great potential business benefits. We publicly release FS-G, as submodules of FederatedScope, at https://github.com/alibaba/FederatedScope to promote FGL's research and enable broad applications that would otherwise be infeasible due to the lack of a dedicated package.

Daoyuan Chen, Dawei Gao, Weirui Kuang et al.

Personalized Federated Learning (pFL), which utilizes and deploys distinct local models, has gained increasing attention in recent years due to its success in handling the statistical heterogeneity of FL clients. However, standardized evaluation and systematical analysis of diverse pFL methods remain a challenge. Firstly, the highly varied datasets, FL simulation settings and pFL implementations prevent easy and fair comparisons of pFL methods. Secondly, the current pFL literature diverges in the adopted evaluation and ablation protocols. Finally, the effectiveness and robustness of pFL methods are under-explored in various practical scenarios, such as the generalization to new clients and the participation of resource-limited clients. To tackle these challenges, we propose the first comprehensive pFL benchmark, pFL-Bench, for facilitating rapid, reproducible, standardized and thorough pFL evaluation. The proposed benchmark contains more than 10 dataset variants in various application domains with a unified data partition and realistic heterogeneous settings; a modularized and easy-to-extend pFL codebase with more than 20 competitive pFL method implementations; and systematic evaluations under containerized environments in terms of generalization, fairness, system overhead, and convergence. We highlight the benefits and potential of state-of-the-art pFL methods and hope the pFL-Bench enables further pFL research and broad applications that would otherwise be difficult owing to the absence of a dedicated benchmark. The code is released at https://github.com/alibaba/FederatedScope/tree/master/benchmark/pFL-Bench.

Daoyuan Chen, Yilun Huang, Zhijian Ma et al.

The immense evolution in Large Language Models (LLMs) has underscored the importance of massive, heterogeneous, and high-quality data. A data recipe is a mixture of data from different sources for training LLMs, which plays a vital role in LLMs' performance. Existing open-source tools for LLM data processing are mostly tailored for specific data recipes. To continuously uncover the potential of LLMs, incorporate data from new sources, and improve LLMs' performance, we build a new system named Data-Juicer, with which we can efficiently generate diverse data recipes, explore different possibilities in forming data mixtures, and evaluate their effects on model performance. Different from traditional data-analytics pipelines, Data-Juicer faces some unique challenges. Firstly, the possible data sources for forming data recipes are truly heterogeneous and massive with various qualities. Secondly, it is extremely expensive to precisely evaluate data recipes' impact on LLMs' performance. Thirdly, the end users of Data-Juicer, model developers, need sufficient flexibility to configure and evaluate different data recipes. Data-Juicer features a fine-grained abstraction of pipelines for constructing data recipes, with over 50 built-in operators for easy composition and extension. By incorporating visualization and auto-evaluation capabilities, Data-Juicer enables a timely feedback loop for both LLM pre-training and fine-tuning. Further, Data-Juicer is optimized and integrated with ecosystems for LLM training, evaluation, and distributed computing. The data recipes derived with Data-Juicer gain notable improvements on state-of-the-art LLMs, by up to 7.45% increase in averaged score across 16 LLM benchmarks and 17.5% higher win rate in pair-wise GPT-4 evaluations. Our system, data recipes, and tutorials are released, calling for broader data-centric research on training and understanding LLMs.

Zeyu Qin, Liuyi Yao, Daoyuan Chen et al.

In this work, besides improving prediction accuracy, we study whether personalization could bring robustness benefits to backdoor attacks. We conduct the first study of backdoor attacks in the pFL framework, testing 4 widely used backdoor attacks against 6 pFL methods on benchmark datasets FEMNIST and CIFAR-10, a total of 600 experiments. The study shows that pFL methods with partial model-sharing can significantly boost robustness against backdoor attacks. In contrast, pFL methods with full model-sharing do not show robustness. To analyze the reasons for varying robustness performances, we provide comprehensive ablation studies on different pFL methods. Based on our findings, we further propose a lightweight defense method, Simple-Tuning, which empirically improves defense performance against backdoor attacks. We believe that our work could provide both guidance for pFL application in terms of its robustness and offer valuable insights to design more robust FL methods in the future. We open-source our code to establish the first benchmark for black-box backdoor attacks in pFL: https://github.com/alibaba/FederatedScope/tree/backdoor-bench.

Liuyi Yao, Dawei Gao, Zhen Wang et al.

To investigate the heterogeneity in federated learning in real-world scenarios, we generalize the classic federated learning to federated hetero-task learning, which emphasizes the inconsistency across the participants in federated learning in terms of both data distribution and learning tasks. We also present B-FHTL, a federated hetero-task learning benchmark consisting of simulation dataset, FL protocols and a unified evaluation mechanism. B-FHTL dataset contains three well-designed federated learning tasks with increasing heterogeneity. Each task simulates the clients with different non-IID data and learning tasks. To ensure fair comparison among different FL algorithms, B-FHTL builds in a full suite of FL protocols by providing high-level APIs to avoid privacy leakage, and presets most common evaluation metrics spanning across different learning tasks, such as regression, classification, text generation and etc. Furthermore, we compare the FL algorithms in fields of federated multi-task learning, federated personalization and federated meta learning within B-FHTL, and highlight the influence of heterogeneity and difficulties of federated hetero-task learning. Our benchmark, including the federated dataset, protocols, the evaluation mechanism and the preliminary experiment, is open-sourced at https://github.com/alibaba/FederatedScope/tree/master/benchmark/B-FHTL

Junkang Wu, Yuexiang Xie, Zhengyi Yang et al.

Direct Preference Optimization (DPO) has emerged as a compelling approach for training Large Language Models (LLMs) to adhere to human preferences. However, the performance of DPO is sensitive to the fine-tuning of its trade-off parameter $β$, as well as to the quality of the preference data. We analyze the impact of $β$ and data quality on DPO, uncovering that optimal $β$ values vary with the informativeness of pairwise data. Addressing the limitations of static $β$ values, we introduce a novel framework that dynamically calibrates $β$ at the batch level, informed by data quality considerations. Additionally, our method incorporates $β$-guided data filtering to safeguard against the influence of outliers. Through empirical evaluation, we demonstrate that our dynamic $β$ adjustment technique significantly improves DPO's performance across a range of models and datasets, offering a more robust and adaptable training paradigm for aligning LLMs with human feedback. The code is available at \url{https://github.com/junkangwu/beta-DPO}.

Chenhe Dong, Yuexiang Xie, Bolin Ding et al.

The increasing privacy concerns on personal private text data promote the development of federated learning (FL) in recent years. However, the existing studies on applying FL in NLP are not suitable to coordinate participants with heterogeneous or private learning objectives. In this study, we further broaden the application scope of FL in NLP by proposing an Assign-Then-Contrast (denoted as ATC) framework, which enables clients with heterogeneous NLP tasks to construct an FL course and learn useful knowledge from each other. Specifically, the clients are suggested to first perform local training with the unified tasks assigned by the server rather than using their own learning objectives, which is called the Assign training stage. After that, in the Contrast training stage, clients train with different local learning objectives and exchange knowledge with other clients who contribute consistent and useful model updates. We conduct extensive experiments on six widely-used datasets covering both Natural Language Understanding (NLU) and Natural Language Generation (NLG) tasks, and the proposed ATC framework achieves significant improvements compared with various baseline methods. The source code is available at \url{https://github.com/alibaba/FederatedScope/tree/master/federatedscope/nlp/hetero_tasks}.

Junkang Wu, Yuexiang Xie, Zhengyi Yang et al.

This study addresses the challenge of noise in training datasets for Direct Preference Optimization (DPO), a method for aligning Large Language Models (LLMs) with human preferences. We categorize noise into pointwise noise, which includes low-quality data points, and pairwise noise, which encompasses erroneous data pair associations that affect preference rankings. Utilizing Distributionally Robust Optimization (DRO), we enhance DPO's resilience to these types of noise. Our theoretical insights reveal that DPO inherently embeds DRO principles, conferring robustness to pointwise noise, with the regularization coefficient $β$ playing a critical role in its noise resistance. Extending this framework, we introduce Distributionally Robustifying DPO (Dr. DPO), which integrates pairwise robustness by optimizing against worst-case pairwise scenarios. The novel hyperparameter $β'$ in Dr. DPO allows for fine-tuned control over data pair reliability, providing a strategic balance between exploration and exploitation in noisy training environments. Empirical evaluations demonstrate that Dr. DPO substantially improves the quality of generated text and response accuracy in preference datasets, showcasing enhanced performance in both noisy and noise-free settings. The code is available at https://github.com/junkangwu/Dr_DPO.

Chiyu Ma, Shuo Yang, Kexin Huang et al.

We present Future-KL Influenced Policy Optimization (FIPO), a reinforcement learning algorithm designed to overcome reasoning bottlenecks in large language models. While GRPO style training scales effectively, it typically relies on outcome-based rewards (ORM) that distribute a global advantage uniformly across every token in a trajectory. We argue that this coarse-grained credit assignment imposes a performance ceiling by failing to distinguish critical logical pivots from trivial tokens. FIPO addresses this by incorporating discounted future-KL divergence into the policy update, creating a dense advantage formulation that re-weights tokens based on their influence on subsequent trajectory behavior. Empirically, FIPO enables models to break through the length stagnation seen in standard baselines. Evaluated on Qwen2.5-32B, FIPO extends the average chain-of-thought length from roughly 4,000 to over 10,000 tokens and increases AIME 2024 Pass@1 accuracy from 50.0% to a peak of 58.0% (converging at approximately 56.0\%). This outperforms both DeepSeek-R1-Zero-Math-32B (around 47.0%) and o1-mini (approximately 56.0%). Our results suggest that establishing dense advantage formulations is a vital path for evolving ORM-based algorithms to unlock the full reasoning potential of base models. We open-source our training system, built on the verl framework.

Yuzheng Hu, Fan Wu, Qinbin Li et al.

As the prevalence of data analysis grows, safeguarding data privacy has become a paramount concern. Consequently, there has been an upsurge in the development of mechanisms aimed at privacy-preserving data analyses. However, these approaches are task-specific; designing algorithms for new tasks is a cumbersome process. As an alternative, one can create synthetic data that is (ideally) devoid of private information. This paper focuses on privacy-preserving data synthesis (PPDS) by providing a comprehensive overview, analysis, and discussion of the field. Specifically, we put forth a master recipe that unifies two prominent strands of research in PPDS: statistical methods and deep learning (DL)-based methods. Under the master recipe, we further dissect the statistical methods into choices of modeling and representation, and investigate the DL-based methods by different generative modeling principles. To consolidate our findings, we provide comprehensive reference tables, distill key takeaways, and identify open problems in the existing literature. In doing so, we aim to answer the following questions: What are the design principles behind different PPDS methods? How can we categorize these methods, and what are the advantages and disadvantages associated with each category? Can we provide guidelines for method selection in different real-world scenarios? We proceed to benchmark several prominent DL-based methods on the task of private image synthesis and conclude that DP-MERF is an all-purpose approach. Finally, upon systematizing the work over the past decade, we identify future directions and call for actions from researchers.

Qingxu Fu, Boyin Liu, Shuchang Tao et al.

We present AgentJet, a distributed swarm training framework for large language model (LLM) agent reinforcement learning. Unlike centralized frameworks that tightly couple agent rollouts with model optimization, AgentJet adopts a decoupled multi-node architecture in which swarm server nodes host trainable models and run optimization on GPU clusters, whereas swarm client nodes execute arbitrary agents on arbitrary devices. This design provides capabilities that are difficult to support in centralized frameworks: (1) heterogeneous multi-model reinforcement learning, enabling the training of heterogeneous multi-agent teams with multiple LLM as brains; (2) multi-task cocktail training with isolated agent runtimes; (3) fault-tolerant execution that prevents external environment failures from interrupting the training process; and (4) live code iteration, which allows agents to be edited during training by replacing swarm client nodes. To support efficient RL in multi-model, multi-turn, and multi-agent settings, AgentJet introduces a context tracking module with timeline merging, which consolidates redundant context and achieves a 1.5-10x training speedup. Finally, AgentJet introduces an automated research system that takes a research topic as input and autonomously conducts long-horizon, multi-day RL studies on large-scale clusters. By leveraging the swarm architecture, this system reproduces key exploratory workflows of RL researchers without human intervention during execution.

Zhen Qin, Daoyuan Chen, Wenhao Zhang et al.

The rapid development of large language models (LLMs) has been witnessed in recent years. Based on the powerful LLMs, multi-modal LLMs (MLLMs) extend the modality from text to a broader spectrum of domains, attracting widespread attention due to the broader range of application scenarios. As LLMs and MLLMs rely on vast amounts of model parameters and data to achieve emergent capabilities, the importance of data is receiving increasingly widespread attention and recognition. Tracing and analyzing recent data-oriented works for MLLMs, we find that the development of models and data is not two separate paths but rather interconnected. On the one hand, vaster and higher-quality data contribute to better performance of MLLMs; on the other hand, MLLMs can facilitate the development of data. The co-development of multi-modal data and MLLMs requires a clear view of 1) at which development stages of MLLMs specific data-centric approaches can be employed to enhance certain MLLM capabilities, and 2) how MLLMs, utilizing those capabilities, can contribute to multi-modal data in specific roles. To promote the data-model co-development for MLLM community, we systematically review existing works related to MLLMs from the data-model co-development perspective. A regularly maintained project associated with this survey is accessible at https://github.com/modelscope/data-juicer/blob/main/docs/awesome_llm_data.md.

Rong Zhu, Wei Chen, Bolin Ding et al.

A recent line of works apply machine learning techniques to assist or rebuild cost-based query optimizers in DBMS. While exhibiting superiority in some benchmarks, their deficiencies, e.g., unstable performance, high training cost, and slow model updating, stem from the inherent hardness of predicting the cost or latency of execution plans using machine learning models. In this paper, we introduce a learning-to-rank query optimizer, called Lero, which builds on top of a native query optimizer and continuously learns to improve the optimization performance. The key observation is that the relative order or rank of plans, rather than the exact cost or latency, is sufficient for query optimization. Lero employs a pairwise approach to train a classifier to compare any two plans and tell which one is better. Such a binary classification task is much easier than the regression task to predict the cost or latency, in terms of model efficiency and accuracy. Rather than building a learned optimizer from scratch, Lero is designed to leverage decades of wisdom of databases and improve the native query optimizer. With its non-intrusive design, Lero can be implemented on top of any existing DBMS with minimal integration efforts. We implement Lero and demonstrate its outstanding performance using PostgreSQL. In our experiments, Lero achieves near optimal performance on several benchmarks. It reduces the plan execution time of the native optimizer in PostgreSQL by up to 70% and other learned query optimizers by up to 37%. Meanwhile, Lero continuously learns and automatically adapts to query workloads and changes in data.

Xuchen Pan, Dawei Gao, Yuexiang Xie et al.

Recent advances in large language models (LLMs) have opened new avenues for applying multi-agent systems in very large-scale simulations. However, there remain several challenges when conducting multi-agent simulations with existing platforms, such as limited scalability and low efficiency, unsatisfied agent diversity, and effort-intensive management processes. To address these challenges, we develop several new features and components for AgentScope, a user-friendly multi-agent platform, enhancing its convenience and flexibility for supporting very large-scale multi-agent simulations. Specifically, we propose an actor-based distributed mechanism as the underlying technological infrastructure towards great scalability and high efficiency, and provide flexible environment support for simulating various real-world scenarios, which enables parallel execution of multiple agents, automatic workflow conversion for distributed deployment, and both inter-agent and agent-environment interactions. Moreover, we integrate an easy-to-use configurable tool and an automatic background generation pipeline in AgentScope, simplifying the process of creating agents with diverse yet detailed background settings. Last but not least, we provide a web-based interface for conveniently monitoring and managing a large number of agents that might deploy across multiple devices. We conduct a comprehensive simulation to demonstrate the effectiveness of these proposed enhancements in AgentScope, and provide detailed observations and insightful discussions to highlight the great potential of applying multi-agent systems in large-scale simulations. The source code is released on GitHub at https://github.com/modelscope/agentscope/tree/main/examples/paper_large_scale_simulation to inspire further research and development in large-scale multi-agent simulations.

Peiyu Liu, Zikang Liu, Ze-Feng Gao et al.

Despite the superior performance, Large Language Models~(LLMs) require significant computational resources for deployment and use. To overcome this issue, quantization methods have been widely applied to reduce the memory footprint of LLMs as well as increasing the inference rate. However, a major challenge is that low-bit quantization methods often lead to performance degradation. It is important to understand how quantization impacts the capacity of LLMs. Different from previous studies focused on overall performance, this work aims to investigate the impact of quantization on \emph{emergent abilities}, which are important characteristics that distinguish LLMs from small language models. Specially, we examine the abilities of in-context learning, chain-of-thought reasoning, and instruction-following in quantized LLMs. Our empirical experiments show that these emergent abilities still exist in 4-bit quantization models, while 2-bit models encounter severe performance degradation on the test of these abilities. To improve the performance of low-bit models, we conduct two special experiments: (1) fine-gained impact analysis that studies which components (or substructures) are more sensitive to quantization, and (2) performance compensation through model fine-tuning. Our work derives a series of important findings to understand the impact of quantization on emergent abilities, and sheds lights on the possibilities of extremely low-bit quantization for LLMs.

Chenhe Dong, Yuexiang Xie, Bolin Ding et al.

Federated learning (FL) enables multiple participants to collaboratively train machine learning models using decentralized data sources, alleviating privacy concerns that arise from directly sharing local data. However, the lack of model privacy protection in FL becomes an unneglectable challenge, especially when people want to federally finetune models based on a proprietary large language model. In this study, we propose a novel FL training approach that accomplishes information exchange among participants via tunable soft prompts. These soft prompts, updated and transmitted between the server and clients, assume the role of the global model parameters and serve as messengers to deliver useful knowledge from the local data and global model. As the global model itself is not required to be shared and the local training is conducted based on an auxiliary model with fewer parameters than the global model, the proposed approach provides protection for the global model while reducing communication and computation costs in FL. Extensive experiments show the effectiveness of the proposed approach compared to several baselines. We have released the source code at \url{https://github.com/alibaba/FederatedScope/tree/fedsp/federatedscope/nlp/fedsp}.

Daoyuan Chen, Haibin Wang, Yilun Huang et al.

The emergence of multimodal large models has advanced artificial intelligence, introducing unprecedented levels of performance and functionality. However, optimizing these models remains challenging due to historically isolated paths of model-centric and data-centric developments, leading to suboptimal outcomes and inefficient resource utilization. In response, we present a new sandbox suite tailored for integrated data-model co-development. This sandbox provides a feedback-driven experimental platform, enabling cost-effective iteration and guided refinement of both data and models. Our proposed ``Probe-Analyze-Refine'' workflow, validated through practical use cases on multimodal tasks such as image-text pre-training with CLIP, image-to-text generation with LLaVA-like models, and text-to-video generation with DiT-based models, yields transferable and notable performance boosts, such as topping the VBench leaderboard. A comprehensive set of over 100 experiments demonstrated the suite's usability and extensibility, while also uncovering insights into the interplay between data quality, diversity, model behavior, and computational costs. All codes, datasets, and models are open-sourced to foster future research and applications that would otherwise be infeasible due to the lack of a dedicated co-development infrastructure.

Runlin Lei, Zhen Wang, Yaliang Li et al.

Graph Neural Networks (GNNs) have received extensive research attention for their promising performance in graph machine learning. Despite their extraordinary predictive accuracy, existing approaches, such as GCN and GPRGNN, are not robust in the face of homophily changes on test graphs, rendering these models vulnerable to graph structural attacks and with limited capacity in generalizing to graphs of varied homophily levels. Although many methods have been proposed to improve the robustness of GNN models, most of these techniques are restricted to the spatial domain and employ complicated defense mechanisms, such as learning new graph structures or calculating edge attentions. In this paper, we study the problem of designing simple and robust GNN models in the spectral domain. We propose EvenNet, a spectral GNN corresponding to an even-polynomial graph filter. Based on our theoretical analysis in both spatial and spectral domains, we demonstrate that EvenNet outperforms full-order models in generalizing across homophilic and heterophilic graphs, implying that ignoring odd-hop neighbors improves the robustness of GNNs. We conduct experiments on both synthetic and real-world datasets to demonstrate the effectiveness of EvenNet. Notably, EvenNet outperforms existing defense models against structural attacks without introducing additional computational costs and maintains competitiveness in traditional node classification tasks on homophilic and heterophilic graphs.

Daoyuan Chen, Dawei Gao, Yuexiang Xie et al.

Federated Learning (FL) aims to train high-quality models in collaboration with distributed clients while not uploading their local data, which attracts increasing attention in both academia and industry. However, there is still a considerable gap between the flourishing FL research and real-world scenarios, mainly caused by the characteristics of heterogeneous devices and its scales. Most existing works conduct evaluations with homogeneous devices, which are mismatched with the diversity and variability of heterogeneous devices in real-world scenarios. Moreover, it is challenging to conduct research and development at scale with heterogeneous devices due to limited resources and complex software stacks. These two key factors are important yet underexplored in FL research as they directly impact the FL training dynamics and final performance, making the effectiveness and usability of FL algorithms unclear. To bridge the gap, in this paper, we propose an efficient and scalable prototyping system for real-world cross-device FL, FS-Real. It supports heterogeneous device runtime, contains parallelism and robustness enhanced FL server, and provides implementations and extensibility for advanced FL utility features such as personalization, communication compression and asynchronous aggregation. To demonstrate the usability and efficiency of FS-Real, we conduct extensive experiments with various device distributions, quantify and analyze the effect of the heterogeneous device and various scales, and further provide insights and open discussions about real-world FL scenarios. Our system is released to help to pave the way for further real-world FL research and broad applications involving diverse devices and scales.

Yunxiang Su, Tianjing Zeng, Zhongjun Ding et al.

With the development of large language models (LLMs), numerous studies integrate LLMs through operator-like components to enhance relational data processing tasks, e.g., filters with semantic predicates, knowledge-augmented table imputation, reasoning-driven entity matching and more challenging semantic query processing. These components invoke LLMs while preserving a relational input/output interface, which we refer to as LLM-Enhanced Relational Operators (LROs). From an operator perspective, unfortunately, these existing LROs suffer from fragmented definition, various implementation strategies and inadequate evaluation benchmarks. To this end, in this paper, we first establish a unified LRO taxonomy to align existing LROs, and categorize them into: Select, Match, Impute, Cluster and Order, along with their operands and implementation variants. Second, we design LROBench, a comprehensive benchmark featuring 290 single-LRO queries and 60 multi-LRO queries, spanning 27 databases across more than 10 domains. LROBench covers all operating logics and operand granularities in its single-LRO workload, and provides challenging multi-LRO queries stratified by query complexity. Based on these, we evaluate individual LROs under various implementations, deriving practical insights into LRO design choices and summarizing our empirical best practices. We further compare the end-to-end performance of existing multi-LRO systems against an LRO suite instantiated with these best practices, in order to investigate how to design an effective LRO set for multi-LRO systems targeting complex semantic queries. Last, to facilitate future work, we outline promising future directions and open-source all benchmark data and evaluation code, available at https://github.com/LROBench/LROBench/.

Fangyuan Zhao, Yuexiang Xie, Xuebin Ren et al.

Federated learning (FL) becomes vulnerable to Byzantine attacks where some of participators tend to damage the utility or discourage the convergence of the learned model via sending their malicious model updates. Previous works propose to apply robust rules to aggregate updates from participators against different types of Byzantine attacks, while at the same time, attackers can further design advanced Byzantine attack algorithms targeting specific aggregation rule when it is known. In practice, FL systems can involve a black-box server that makes the adopted aggregation rule inaccessible to participants, which can naturally defend or weaken some Byzantine attacks. In this paper, we provide an in-depth understanding on the Byzantine robustness of the FL system with a black-box server. Our investigation demonstrates the improved Byzantine robustness of a black-box server employing a dynamic defense strategy. We provide both empirical evidence and theoretical analysis to reveal that the black-box server can mitigate the worst-case attack impact from a maximum level to an expectation level, which is attributed to the inherent inaccessibility and randomness offered by a black-box server.The source code is available at https://github.com/alibaba/FederatedScope/tree/Byzantine_attack_defense to promote further research in the community.

Haoming Meng, Kexin Huang, Shaohang Wei et al. · pku

Reinforcement learning with verifiable rewards (RLVR) has significantly improved reasoning in large language models (LLMs), yet the token-level mechanisms underlying these improvements remain unclear. We present a systematic empirical study of RLVR's distributional effects organized around three main analyses: (1) token-level characterization of distributional shifts between base and RL models, (2) the impact of token-level distributional shifts on sequence-level reasoning performance through cross-sampling interventions, and (3) fine-grained mechanics of these shifts at the token level. We find that RL fine-tuning induces highly sparse and targeted changes, with only a small fraction of token distributions exhibiting meaningful divergence between the base and RL policies. We further characterize the structure and evolution of these shifts through analyses of token entropy, positional concentration, and reallocation of probability mass. To assess the functional importance of these sparse changes, we conduct cross-sampling experiments that selectively swap token choices between the base and RL models with varying intervention budgets. We show that inserting only a small fraction of RL-sampled tokens into base generations progressively recovers RL performance gains, while injecting a similarly small number of base token choices into otherwise RL-generated sequences collapses performance to base levels, isolating a small set of token-level decisions directly responsible for RLVR's performance gains. Finally, we explore divergence-weighted variants of the advantage signal as a diagnostic intervention, finding that they can yield improvements over baselines. Together, our results shed light on the distributional changes induced by RLVR and provide a fine-grained, token-level lens for understanding RLVR fine-tuning as a targeted refinement process.

Yige Yuan, Teng Xiao, Shuchang Tao et al.

Large language models (LLMs) have demonstrated impressive performance on reasoning-intensive tasks, but enhancing their reasoning abilities typically relies on either reinforcement learning (RL) with verifiable signals or supervised fine-tuning (SFT) with high-quality long chain-of-thought (CoT) demonstrations, both of which are expensive. In this paper, we study a novel problem of incentivizing the reasoning capacity of LLMs without expensive high-quality demonstrations and reinforcement learning. We investigate whether the reasoning capabilities of LLMs can be effectively incentivized via supervision from significantly weaker models. We further analyze when and why such weak supervision succeeds in eliciting reasoning abilities in stronger models. Our findings show that supervision from significantly weaker reasoners can substantially improve student reasoning performance, recovering close to 94% of the gains of expensive RL at a fraction of the cost. Experiments across diverse benchmarks and model architectures demonstrate that weak reasoners can effectively incentivize reasoning in stronger student models, consistently improving performance across a wide range of reasoning tasks. Our results suggest that this simple weak-to-strong paradigm is a promising and generalizable alternative to costly methods for incentivizing strong reasoning capabilities at inference-time in LLMs. The code is publicly available at https://github.com/yuanyige/w2sr.

Yuping Lin, Zitao Li, Yue Xing et al.

Recent studies have identified "retrieval heads" in Large Language Models (LLMs) responsible for extracting information from input contexts. However, prior works largely rely on static statistics aggregated across datasets, identifying heads that perform retrieval on average. This perspective overlooks the fine-grained temporal dynamics of autoregressive generation. In this paper, we investigate retrieval heads from a dynamic perspective. Through extensive analysis, we establish three core claims: (1) Dynamism: Retrieval heads vary dynamically across timesteps; (2) Irreplaceability: Dynamic retrieval heads are specific at each timestep and cannot be effectively replaced by static retrieval heads; and (3) Correlation: The model's hidden state encodes a predictive signal for future retrieval head patterns, indicating an internal planning mechanism. We validate these findings on the Needle-in-a-Haystack task and a multi-hop QA task, and quantify the differences on the utility of dynamic and static retrieval heads in a Dynamic Retrieval-Augmented Generation framework. Our study provides new insights into the internal mechanisms of LLMs.

Lingzhe Zhang, Yunpeng Zhai, Tong Jia et al.

Large Language Models (LLMs) integrated with agent-based reasoning frameworks have recently shown strong potential for autonomous decision-making and system-level operations. One promising yet underexplored direction is microservice remediation, where the goal is to automatically recover faulty microservice systems. Existing approaches, however, still rely on human-crafted prompts from Site Reliability Engineers (SREs), with LLMs merely converting textual instructions into executable code. To advance research in this area, we introduce MicroRemed, the first benchmark for evaluating LLMs in end-to-end microservice remediation, where models must directly generate executable Ansible playbooks from diagnosis reports to restore system functionality. We further propose ThinkRemed, a multi-agent framework that emulates the reflective and perceptive reasoning of SREs. Experimental results show that MicroRemed presents substantial challenges to current LLMs, while ThinkRemed improves end-to-end remediation performance through iterative reasoning and system reflection. The benchmark is available at https://github.com/LLM4AIOps/MicroRemed.

Qirui Jiao, Daoyuan Chen, Yilun Huang et al.

High-performance Multimodal Large Language Models (MLLMs) are heavily dependent on data quality. To advance fine-grained image recognition within MLLMs, we introduce a novel data synthesis method inspired by contrastive learning and image difference captioning. Our key idea involves challenging the model to discern both matching and distinct elements by scrutinizing object differences in detailed regions across similar images. We begin by generating pairs of similar images that emphasize object variations. Following this, we employ a Difference Area Generator to pinpoint object differences, and subsequently, a Difference Captions Generator to articulate these differences. This process results in a high-quality dataset of "object replacement" samples, termed Img-Diff, which can be scaled as needed due to its automated nature. We leverage this generated dataset to fine-tune state-of-the-art (SOTA) MLLMs, such as InternVL2, achieving substantial improvements across various image difference and Visual Question Answering tasks. Notably, the trained models significantly outperform existing SOTA models like GPT-4V and Gemini on the MMVP benchmark. Additionally, we conduct comprehensive evaluations to validate the dataset's diversity, quality, and robustness, offering several insights into the synthesis of such contrastive datasets. We release our codes and dataset to encourage further research on multimodal data synthesis and MLLMs' fundamental capabilities for image understanding.

Yuntao Du, Zitao Li, Bolin Ding et al.

Impressive progress has been made in automated problem-solving by the collaboration of large language model (LLM) based agents. However, these automated capabilities also open avenues for malicious applications. In this paper, we study a new threat that LLMs pose to online pseudonymity, called automated profile inference, where an adversary can instruct LLMs to automatically collect and extract sensitive personal attributes from publicly available user activities on pseudonymous platforms. We also introduce an automated profiling framework called AutoProfiler to demonstrate and assess the feasibility of such attacks in real-world scenarios. AutoProfiler consists of four specialized LLM agents that work collaboratively to retrieve and process user online activities and generate a profile with extracted personal information. Experimental results on two real-world datasets and one synthetic dataset show that AutoProfiler is highly effective and efficient, and the inferred attributes are both identifiable and sensitive, posing significant privacy risks. We explore mitigation strategies from different perspectives and advocate for increased public awareness of this emerging privacy threat.

Shinji Mai, Yunpeng Zhai, Ziqian Chen et al.

Large language model based agents are increasingly deployed in complex, tool augmented environments. While reinforcement learning provides a principled mechanism for such agents to improve through interaction, its effectiveness critically depends on the availability of structured training tasks. In many realistic settings, however, no such tasks exist a challenge we term task scarcity, which has become a key bottleneck for scaling agentic RL. Existing approaches typically assume predefined task collections, an assumption that fails in novel environments where tool semantics and affordances are initially unknown. To address this limitation, we formalize the problem of Task Generation for Agentic RL, where an agent must learn within a given environment that lacks predefined tasks. We propose CuES, a Curiosity driven and Environment grounded Synthesis framework that autonomously generates diverse, executable, and meaningful tasks directly from the environment structure and affordances, without relying on handcrafted seeds or external corpora. CuES drives exploration through intrinsic curiosity, abstracts interaction patterns into reusable task schemas, and refines them through lightweight top down guidance and memory based quality control. Across three representative environments, AppWorld, BFCL, and WebShop, CuES produces task distributions that match or surpass manually curated datasets in both diversity and executability, yielding substantial downstream policy improvements. These results demonstrate that curiosity driven, environment grounded task generation provides a scalable foundation for agents that not only learn how to act, but also learn what to learn. The code is available at https://github.com/modelscope/AgentEvolver/research/CuES.

Yanxi Chen, Yaliang Li, Bolin Ding et al.

This work presents an analytical framework for the design and analysis of LLM-based algorithms, i.e., algorithms that contain one or multiple calls of large language models (LLMs) as sub-routines and critically rely on the capabilities of LLMs. While such algorithms, ranging from basic LLM calls with prompt engineering to complicated LLM-powered agentic workflows and compound AI systems, have achieved remarkable empirical success, their design and optimization oftentimes require extensive trial-and-errors and case-by-case analysis. Our proposed framework serves as an attempt to mitigate such headaches, offering a formal and systematic approach for analyzing how the accuracy and efficiency of an LLM-based algorithm will be impacted by critical design choices, such as the pattern and granularity of task decomposition, or the prompt for each LLM call. Through a wide range of case studies covering diverse algorithm patterns (including parallel/hierarchical/recursive task decomposition and generic directed acyclic graphs), we demonstrate the proposed framework in action and derive interesting insights that generalize across scenarios, accompanied by systematic empirical validation in synthetic settings.

Kexin Huang, Haoming Meng, Junkang Wu et al.

Reinforcement learning with verifiable rewards (RLVR) has substantially improved the reasoning capabilities of large language models. While existing analyses identify that RLVR-induced changes are sparse, they primarily focus on the \textbf{magnitude} of these updates, largely overlooking their \textbf{direction}. In this work, we argue that the direction of updates is a more critical lens for understanding RLVR's effects, which can be captured by the signed, token-level log probability difference $Δ\log p$ between the base and final RLVR models. Through statistical analysis and token-replacement interventions, we demonstrate that $Δ\log p$ more effectively identifies sparse, yet reasoning-critical updates than magnitude-based metrics (\eg divergence or entropy). Building on this insight, we propose two practical applications: (1) a \textit{test-time extrapolation} method that amplifies the policy along the learned $Δ\log p$ direction to improve reasoning accuracy without further training; (2) a \textit{training-time reweighting} method that focuses learning on low-probability (corresponding to higher $Δ\log p$) tokens, which improves reasoning performance across models and benchmarks. Our work establishes the direction of change as a key principle for analyzing and improving RLVR.

Yuchang Sun, Yuexiang Xie, Bolin Ding et al.

Federated learning (FL) has emerged as a promising paradigm for fine-tuning foundation models using distributed data in a privacy-preserving manner. Under limited computational resources, clients often find it more practical to fine-tune a selected subset of layers, rather than the entire model, based on their task-specific data. In this study, we provide a thorough theoretical exploration of selective layer fine-tuning in FL, emphasizing a flexible approach that allows the clients to adjust their selected layers according to their local data and resources. We theoretically demonstrate that the layer selection strategy has a significant impact on model convergence in two critical aspects: the importance of selected layers and the heterogeneous choices across clients. Drawing from these insights, we further propose a strategic layer selection method that utilizes local gradients and regulates layer selections across clients. The extensive experiments on both image and text datasets demonstrate the effectiveness of the proposed strategy compared with several baselines, highlighting its advances in identifying critical layers that adapt to the client heterogeneity and training dynamics in FL.

Dawei Gao, Zitao Li, Xuchen Pan et al.

With the rapid advancement of Large Language Models (LLMs), significant progress has been made in multi-agent applications. However, the complexities in coordinating agents' cooperation and LLMs' erratic performance pose notable challenges in developing robust and efficient multi-agent applications. To tackle these challenges, we propose AgentScope, a developer-centric multi-agent platform with message exchange as its core communication mechanism. The abundant syntactic tools, built-in agents and service functions, user-friendly interfaces for application demonstration and utility monitor, zero-code programming workstation, and automatic prompt tuning mechanism significantly lower the barriers to both development and deployment. Towards robust and flexible multi-agent application, AgentScope provides both built-in and customizable fault tolerance mechanisms. At the same time, it is also armed with system-level support for managing and utilizing multi-modal data, tools, and external knowledge. Additionally, we design an actor-based distribution framework, enabling easy conversion between local and distributed deployments and automatic parallel optimization without extra effort. With these features, AgentScope empowers developers to build applications that fully realize the potential of intelligent agents. We have released AgentScope at https://github.com/modelscope/agentscope, and hope AgentScope invites wider participation and innovation in this fast-moving field.

Yanxi Chen, Xuchen Pan, Yaliang Li et al.

We present EE-LLM, a framework for large-scale training and inference of early-exit large language models (LLMs). While recent works have shown preliminary evidence for the efficacy of early exiting in accelerating LLM inference, EE-LLM makes a foundational step towards scaling up early-exit LLMs by supporting their training and inference with massive 3D parallelism. Built upon Megatron-LM, EE-LLM implements a variety of algorithmic innovations and performance optimizations tailored to early exiting, including a lightweight method that facilitates backpropagation for the early-exit training objective with pipeline parallelism, techniques of leveraging idle resources in the original pipeline schedule for computation related to early-exit layers, and two approaches of early-exit inference that are compatible with KV caching for autoregressive generation. Our analytical and empirical study shows that EE-LLM achieves great training efficiency with negligible computational overhead compared to standard LLM training, as well as outstanding inference speedup without compromising output quality. To facilitate further research and adoption, we release EE-LLM at https://github.com/pan-x-c/EE-LLM.

Lingzhe Zhang, Yunpeng Zhai, Tong Jia et al.

Contemporary microservice systems continue to grow in scale and complexity, leading to increasingly frequent and costly failures. While recent LLM-based auto-remediation approaches have emerged, they primarily translate textual instructions into executable Ansible playbooks and rely on expert-crafted prompts, lacking runtime knowledge guidance and depending on large-scale general-purpose LLMs, which limits their accuracy and efficiency. We introduce \textit{End-to-End Microservice Remediation} (E2E-MR), a new task that requires directly generating executable playbooks from diagnosis reports to autonomously restore faulty systems. To enable rigorous evaluation, we build \textit{MicroRemed}, a benchmark that automates microservice deployment, failure injection, playbook execution, and post-repair verification. We further propose \textit{E2E-REME}, an end-to-end auto-remediation model trained via experience-simulation reinforcement fine-tuning. Experiments on public and industrial microservice platforms, compared with nine representative LLMs, show that E2E-REME achieves superior accuracy and efficiency.

Shuo Yang, Jinda Lu, Chiyu Ma et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a central paradigm for scaling LLM reasoning, yet its optimization often suffers from training instability and suboptimal convergence. Through a systematic dissection of clipping-based GRPO-style objectives, we identify the rigid clipping decision induced by hard clipping as a key practical bottleneck in the studied RLVR setups. Specifically, our analysis suggests that informative signals can lie in the near-boundary region just beyond the clipping threshold, and are therefore discarded by the standard hard-clipping rule. Notably, once this bottleneck is precisely identified, even simple stochastic perturbations at the boundary can recover meaningful performance gains. Building on this finding, we propose Near-boundary Stochastic Rescue (NSR), a minimal, plug-and-play modification that stochastically retains these slightly out-of-bound tokens to recover lost signals. While NSR, via stochastic sampling, can be interpreted as inducing an implicit gradient decay in expectation, our ablations reveal that its stochastic, boundary-local rescue mechanism is consistently more effective than deterministic gradient decay. Validated by extensive experiments across model sizes from 7B to 30B and both dense and MoE architectures, as a plug-and-play solution, NSR substantially improves training stability and delivers consistent gains over strong baselines such as DAPO and GSPO.

Lizhe Fang, Yifei Wang, Zhaoyang Liu et al.

Handling long-context inputs is crucial for large language models (LLMs) in tasks such as extended conversations, document summarization, and many-shot in-context learning. While recent approaches have extended the context windows of LLMs and employed perplexity (PPL) as a standard evaluation metric, PPL has proven unreliable for assessing long-context capabilities. The underlying cause of this limitation has remained unclear. In this work, we provide a comprehensive explanation for this issue. We find that PPL overlooks key tokens, which are essential for long-context understanding, by averaging across all tokens and thereby obscuring the true performance of models in long-context scenarios. To address this, we propose \textbf{LongPPL}, a novel metric that focuses on key tokens by employing a long-short context contrastive method to identify them. Our experiments demonstrate that LongPPL strongly correlates with performance on various long-context benchmarks (e.g., Pearson correlation of -0.96), significantly outperforming traditional PPL in predictive accuracy. Additionally, we introduce \textbf{LongCE} (Long-context Cross-Entropy) loss, a re-weighting strategy for fine-tuning that prioritizes key tokens, leading to consistent improvements across diverse benchmarks. In summary, these contributions offer deeper insights into the limitations of PPL and present effective solutions for accurately evaluating and enhancing the long-context capabilities of LLMs. Code is available at https://github.com/PKU-ML/LongPPL.

Leyi Pan, Shuchang Tao, Yunpeng Zhai et al.

Reliable reinforcement learning (RL) for diffusion large language models (dLLMs) requires both accurate advantage estimation and precise estimation of prediction probabilities. Existing RL methods for dLLMs fall short in both aspects: they rely on coarse or unverifiable reward signals, and they estimate prediction probabilities without accounting for the bias relative to the true, unbiased expected prediction probability that properly integrates over all possible decoding orders. To mitigate these issues, we propose \emph{d}-TreeRPO, a reliable RL framework for dLLMs that leverages tree-structured rollouts and bottom-up advantage computation based on verifiable outcome rewards to provide fine-grained and verifiable step-wise reward signals. When estimating the conditional transition probability from a parent node to a child node, we theoretically analyze the estimation error between the unbiased expected prediction probability and the estimate obtained via a single forward pass, and find that higher prediction confidence leads to lower estimation error. Guided by this analysis, we introduce a time-scheduled self-distillation loss during training that enhances prediction confidence in later training stages, thereby enabling more accurate probability estimation and improved convergence. Experiments show that \emph{d}-TreeRPO outperforms existing baselines and achieves significant gains on multiple reasoning benchmarks, including +86.2 on Sudoku, +51.6 on Countdown, +4.5 on GSM8K, and +5.3 on Math500. Ablation studies and computational cost analyses further demonstrate the effectiveness and practicality of our design choices.

Wenhao Zhang, Yuexiang Xie, Yuchang Sun et al.

Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) are two prominent post-training paradigms for refining the capabilities and aligning the behavior of Large Language Models (LLMs). Existing approaches that integrate SFT and RL often face the risk of disrupting established response patterns and inducing overfitting to expert data. To address this, we present a novel investigation into the unified view of SFT and RL through an off-policy versus on-policy lens. We propose CHORD, a framework for Controllable Harmonization of On- and Off-Policy Reinforcement Learning via Dynamic Weighting, which reframes SFT not as a separate stage but as a dynamically weighted auxiliary objective within the on-policy RL process. Based on an analysis of off-policy expert data's influence at both holistic and granular levels, we incorporate a dual-control mechanism in CHORD. Specifically, the framework first employs a global coefficient to holistically guide the transition from off-policy imitation to on-policy exploration, and then applies a token-wise weighting function that enables granular learning from the expert, which promotes on-policy exploration and mitigates disruption from off-policy data. We conduct extensive experiments on mathematical reasoning problems and practical tool-use tasks, providing empirical evidence that CHORD achieves a stable and efficient learning process. By effectively harmonizing off-policy expert data with on-policy exploration, CHORD demonstrates significant improvements over baselines. We release the implementation at https://github.com/modelscope/Trinity-RFT/tree/main/examples/mix_chord to inspire further research.

Qianli Shen, Daoyuan Chen, Yilun Huang et al.

Reinforcement finetuning (RFT) is a key technique for aligning Large Language Models (LLMs) with human preferences and enhancing reasoning, yet its effectiveness is highly sensitive to which tasks are explored during training. Uniform task sampling is inefficient, wasting computation on tasks that are either trivial or unsolvable, while existing task selection methods often suffer from high rollout costs, poor adaptivity, or incomplete evidence. We introduce BOTS, a unified framework for Bayesian Online Task Selection in LLM reinforcement finetuning. Grounded in Bayesian inference, BOTS adaptively maintains posterior estimates of task difficulty as the model evolves. It jointly incorporates explicit evidence from direct evaluations of selected tasks and implicit evidence inferred from these evaluations for unselected tasks, with Thompson sampling ensuring a principled balance between exploration and exploitation. To make implicit evidence practical, we instantiate it with an ultra-light interpolation-based plug-in that estimates difficulties of unevaluated tasks without extra rollouts, adding negligible overhead. Empirically, across diverse domains and LLM scales, BOTS consistently improves data efficiency and performance over baselines and ablations, providing a practical and extensible solution for dynamic task selection in RFT.

Ting Zhou, Daoyuan Chen, Qirui Jiao et al.

In the domain of Multimodal Large Language Models (MLLMs), achieving human-centric video understanding remains a formidable challenge. Existing benchmarks primarily emphasize object and action recognition, often neglecting the intricate nuances of human emotions, behaviors, and speech-visual alignment within video content. We present HumanVBench, an innovative benchmark meticulously crafted to bridge these gaps in the evaluation of video MLLMs. HumanVBench comprises 16 carefully designed tasks that explore two primary dimensions: inner emotion and outer manifestations, spanning static and dynamic, basic and complex, as well as single-modal and cross-modal aspects. With two advanced automated pipelines for video annotation and distractor-included QA generation, HumanVBench utilizes diverse state-of-the-art (SOTA) techniques to streamline benchmark data synthesis and quality assessment, minimizing human annotation dependency tailored to human-centric multimodal attributes. A comprehensive evaluation across 22 SOTA video MLLMs reveals notable limitations in current performance, especially in cross-modal and emotion perception, underscoring the necessity for further refinement toward achieving more human-like understanding. HumanVBench is open-sourced to facilitate future advancements and real-world applications in video MLLMs.

Yunpeng Zhai, Shuchang Tao, Cheng Chen et al.

Autonomous agents powered by large language models (LLMs) have the potential to significantly enhance human productivity by reasoning, using tools, and executing complex tasks in diverse environments. However, current approaches to developing such agents remain costly and inefficient, as they typically require manually constructed task datasets and reinforcement learning (RL) pipelines with extensive random exploration. These limitations lead to prohibitively high data-construction costs, low exploration efficiency, and poor sample utilization. To address these challenges, we present AgentEvolver, a self-evolving agent system that leverages the semantic understanding and reasoning capabilities of LLMs to drive autonomous agent learning. AgentEvolver introduces three synergistic mechanisms: (i) self-questioning, which enables curiosity-driven task generation in novel environments, reducing dependence on handcrafted datasets; (ii) self-navigating, which improves exploration efficiency through experience reuse and hybrid policy guidance; and (iii) self-attributing, which enhances sample efficiency by assigning differentiated rewards to trajectory states and actions based on their contribution. By integrating these mechanisms into a unified framework, AgentEvolver enables scalable, cost-effective, and continual improvement of agent capabilities. Preliminary experiments indicate that AgentEvolver achieves more efficient exploration, better sample utilization, and faster adaptation compared to traditional RL-based baselines.

Junkang Wu, Xue Wang, Zhengyi Yang et al.

Aligning large language models (LLMs) with human values and intentions is crucial for their utility, honesty, and safety. Reinforcement learning from human feedback (RLHF) is a popular approach to achieve this alignment, but it faces challenges in computational efficiency and training stability. Recent methods like Direct Preference Optimization (DPO) and Simple Preference Optimization (SimPO) have proposed offline alternatives to RLHF, simplifying the process by reparameterizing the reward function. However, DPO depends on a potentially suboptimal reference model, and SimPO's assumption of a fixed target reward margin may lead to suboptimal decisions in diverse data settings. In this work, we propose $α$-DPO, an adaptive preference optimization algorithm designed to address these limitations by introducing a dynamic reward margin. Specifically, $α$-DPO employs an adaptive preference distribution, balancing the policy model and the reference model to achieve personalized reward margins. We provide theoretical guarantees for $α$-DPO, demonstrating its effectiveness as a surrogate optimization objective and its ability to balance alignment and diversity through KL divergence control. Empirical evaluations on AlpacaEval 2 and Arena-Hard show that $α$-DPO consistently outperforms DPO and SimPO across various model settings, establishing it as a robust approach for fine-tuning LLMs. Our method achieves significant improvements in win rates, highlighting its potential as a powerful tool for LLM alignment. The code is available at https://github.com/junkangwu/alpha-DPO

Xuchen Pan, Yanxi Chen, Yaliang Li et al.

This work introduces EE-Tuning, a lightweight and economical solution to training/tuning early-exit large language models (LLMs). In contrast to the common approach of full-parameter pre-training, EE-Tuning augments any pre-trained (and possibly fine-tuned) standard LLM with additional early-exit layers that are tuned in a parameter-efficient manner, which requires significantly less computational resources and training data. Our implementation of EE-Tuning achieves outstanding training efficiency via extensive performance optimizations, as well as scalability due to its full compatibility with 3D parallelism. Results of systematic experiments validate the efficacy of EE-Tuning, confirming that effective early-exit LLM inference can be achieved with a limited training budget. In hope of making early-exit LLMs accessible to the community, we release the source code of our implementation of EE-Tuning at https://github.com/pan-x-c/EE-LLM.

Jiarui Ji, Runlin Lei, Xuchen Pan et al.

The emergence of Large Language Models (LLMs) demonstrates their potential to encapsulate the logic and patterns inherent in human behavior simulation by leveraging extensive web data pre-training. However, the boundaries of LLM capabilities in social simulation remain unclear. To further explore the social attributes of LLMs, we introduce the CiteAgent framework, designed to generate citation networks based on human-behavior simulation with LLM-based agents. CiteAgent successfully captures predominant phenomena in real-world citation networks, including power-law distribution, citational distortion, and shrinking diameter. Building on this realistic simulation, we establish two LLM-based research paradigms in social science: LLM-SE (LLM-based Survey Experiment) and LLM-LE (LLM-based Laboratory Experiment). These paradigms facilitate rigorous analyses of citation network phenomena, allowing us to validate and challenge existing theories. Additionally, we extend the research scope of traditional science of science studies through idealized social experiments, with the simulation experiment results providing valuable insights for real-world academic environments. Our work demonstrates the potential of LLMs for advancing science of science research in social science.

Xue Wang, Tian Zhou, Jinyang Gao et al.

We present a joint forecasting framework for time series prediction that contrasts with traditional direct or recursive methods. This framework achieves state-of-the-art performance for our designed foundation model, YingLong, and reveals a novel scaling effect: longer outputs significantly enhance model accuracy due to delayed chain-of-thought reasoning in our non-causal approach. YingLong is a non-causal, bidirectional attention encoder-only transformer trained through masked token recovery, aligning more effectively with language understanding tasks than with generation tasks. Additionally, we boost performance by tackling output variance with a multi-input ensemble. We release four foundation models ranging from 6M to 300M parameters, demonstrating superior results in zero-shot tasks on the ETT and Weather datasets. YingLong achieves more than 60% best performance. To ensure generalizability, we assessed the models using the GIFT-Eval benchmark, which comprises 23 time series datasets across 7 domains. Yinglong significantly outperformed the best time-series foundation models, end-to-end trained models by 14% and 44% in rank respectively.The pretrained 300M model is available at https://huggingface.co/qcw1314/YingLong_300m

Jiarui Ji, Runlin Lei, Jialing Bi et al.

The structural properties of naturally arising social graphs are extensively studied to understand their evolution. Prior approaches for modeling network dynamics typically rely on rule-based models, which lack realism and generalizability, or deep learning-based models, which require large-scale training datasets. Social graphs, as abstract graph representations of entity-wise interactions, present an opportunity to explore network evolution mechanisms through realistic simulations of human-item interactions. Leveraging the pre-trained social consensus knowledge embedded in large language models (LLMs), we present GraphAgent-Generator (GAG), a novel simulation-based framework for dynamic, text-attributed social graph generation. GAG simulates the temporal node and edge generation processes for zero-shot social graph generation. The resulting graphs exhibit adherence to seven key macroscopic network properties, achieving an 11% improvement in microscopic graph structure metrics. Through the node classification benchmarking task, we validate GAG effectively captures the intricate text-structure correlations in graph generation. Furthermore, GAG supports generating graphs with up to nearly 100,000 nodes or 10 million edges through large-scale LLM-based agent simulation with parallel acceleration, achieving a minimum speed-up of 90.4%. The source code is available at https://github.com/Ji-Cather/GraphAgent.