Xiaoshuai Song

Runqi Qiao, Qiuna Tan, Guanting Dong et al.

Visual mathematical reasoning, as a fundamental visual reasoning ability, has received widespread attention from the Large Multimodal Models (LMMs) community. Existing benchmarks, such as MathVista and MathVerse, focus more on the result-oriented performance but neglect the underlying principles in knowledge acquisition and generalization. Inspired by human-like mathematical reasoning, we introduce WE-MATH, the first benchmark specifically designed to explore the problem-solving principles beyond end-to-end performance. We meticulously collect and categorize 6.5K visual math problems, spanning 67 hierarchical knowledge concepts and five layers of knowledge granularity. We decompose composite problems into sub-problems according to the required knowledge concepts and introduce a novel four-dimensional metric, namely Insufficient Knowledge (IK), Inadequate Generalization (IG), Complete Mastery (CM), and Rote Memorization (RM), to hierarchically assess inherent issues in LMMs' reasoning process. With WE-MATH, we conduct a thorough evaluation of existing LMMs in visual mathematical reasoning and reveal a negative correlation between solving steps and problem-specific performance. We confirm the IK issue of LMMs can be effectively improved via knowledge augmentation strategies. More notably, the primary challenge of GPT-4o has significantly transitioned from IK to IG, establishing it as the first LMM advancing towards the knowledge generalization stage. In contrast, other LMMs exhibit a marked inclination towards Rote Memorization - they correctly solve composite problems involving multiple knowledge concepts yet fail to answer sub-problems. We anticipate that WE-MATH will open new pathways for advancements in visual mathematical reasoning for LMMs. The WE-MATH data and evaluation code are available at https://github.com/We-Math/We-Math.

Xiaoshuai Song, Haofei Chang, Guanting Dong et al.

Large language models (LLMs) are expected to be trained to act as agents in various real-world environments, but this process relies on rich and varied tool-interaction sandboxes. However, access to real systems is often restricted; LLM-simulated environments are prone to hallucinations and inconsistencies; and manually built sandboxes are hard to scale. In this paper, we propose EnvScaler, an automated framework for scalable tool-interaction environments via programmatic synthesis. EnvScaler comprises two components. First, SkelBuilder constructs diverse environment skeletons through topic mining, logic modeling, and quality evaluation. Then, ScenGenerator generates multiple task scenarios and rule-based trajectory validation functions for each environment. With EnvScaler, we synthesize 191 environments and about 7K scenarios, and apply them to Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) for Qwen3 series models. Results on three benchmarks show that EnvScaler significantly improves LLMs' ability to solve tasks in complex environments involving multi-turn, multi-tool interactions. We release our code and data at https://github.com/RUC-NLPIR/EnvScaler.

Jiachi Liu, Liwen Wang, Guanting Dong et al.

In real dialogue scenarios, as there are unknown input noises in the utterances, existing supervised slot filling models often perform poorly in practical applications. Even though there are some studies on noise-robust models, these works are only evaluated on rule-based synthetic datasets, which is limiting, making it difficult to promote the research of noise-robust methods. In this paper, we introduce a noise robustness evaluation dataset named Noise-SF for slot filling task. The proposed dataset contains five types of human-annotated noise, and all those noises are exactly existed in real extensive robust-training methods of slot filling into the proposed framework. By conducting exhaustive empirical evaluation experiments on Noise-SF, we find that baseline models have poor performance in robustness evaluation, and the proposed framework can effectively improve the robustness of models. Based on the empirical experimental results, we make some forward-looking suggestions to fuel the research in this direction. Our dataset Noise-SF will be released at https://github.com/dongguanting/Noise-SF.

Weixun Wang, XiaoXiao Xu, Wanhe An et al.

Agentic crafting requires LLMs to operate in real-world environments over multiple turns by taking actions, observing outcomes, and iteratively refining artifacts. Despite its importance, the open-source community lacks a principled, end-to-end ecosystem to streamline agent development. We introduce the Agentic Learning Ecosystem (ALE), a foundational infrastructure that optimizes the production pipeline for agentic model. ALE consists of three components: ROLL, a post-training framework for weight optimization; ROCK, a sandbox environment manager for trajectory generation; and iFlow CLI, an agent framework for efficient context engineering. We release ROME, an open-source agent grounded by ALE and trained on over one million trajectories. Our approach includes data composition protocols for synthesizing complex behaviors and a novel policy optimization algorithm, Interaction-Perceptive Agentic Policy Optimization (IPA), which assigns credit over semantic interaction chunks rather than individual tokens to improve long-horizon training stability. Empirically, we evaluate ROME within a structured setting and introduce Terminal Bench Pro, a benchmark with improved scale and contamination control. ROME demonstrates strong performance across benchmarks like SWE-bench Verified and Terminal Bench, proving the effectiveness of ALE.

Pei Wang, Yanan Wu, Xiaoshuai Song et al.

Large language model (LLM)-based agents are increasingly deployed in e-commerce shopping. To perform thorough, user-tailored product searches, agents should interpret personal preferences, engage in multi-turn dialogues, and ultimately retrieve and discriminate among highly similar products. However, existing research has yet to provide a unified simulation environment that consistently captures all of these aspects, and always focuses solely on evaluation benchmarks without training support. In this paper, we introduce ShopSimulator, a large-scale and challenging Chinese shopping environment. Leveraging ShopSimulator, we evaluate LLMs across diverse scenarios, finding that even the best-performing models achieve less than 40% full-success rate. Error analysis reveals that agents struggle with deep search and product selection in long trajectories, fail to balance the use of personalization cues, and to effectively engage with users. Further training exploration provides practical guidance for overcoming these weaknesses, with the combination of supervised fine-tuning (SFT) and reinforcement learning (RL) yielding significant performance improvements. Code and data will be released at https://github.com/ShopAgent-Team/ShopSimulator.

Guanting Dong, Junting Lu, Junjie Huang et al.

Large language models are increasingly expected to serve as general-purpose agents that interact with external, stateful tool environments. The Model Context Protocol (MCP) and broader agent skills offer a unified interface for connecting agents with scalable real-world services, but training robust agents remains limited by the lack of realistic environments and principled mechanisms for life-long learning. In this paper, we present \textbf{Agent-World}, a self-evolving training arena for advancing general agent intelligence through scalable environments. Agent-World has two main components: (1) Agentic Environment-Task Discovery, which autonomously explores topic-aligned databases and executable tool ecosystems from thousands of real-world environment themes and synthesizes verifiable tasks with controllable difficulty; and (2) Continuous Self-Evolving Agent Training, which combines multi-environment reinforcement learning with a self-evolving agent arena that automatically identifies capability gaps through dynamic task synthesis and drives targeted learning, enabling the co-evolution of agent policies and environments. Across 23 challenging agent benchmarks, Agent-World-8B and 14B consistently outperforms strong proprietary models and environment scaling baselines. Further analyses reveal scaling trends in relation to environment diversity and self-evolution rounds, offering insights for building general agent intelligence.

Xiaoshuai Song, Keqing He, Pei Wang et al.

The tasks of out-of-domain (OOD) intent discovery and generalized intent discovery (GID) aim to extend a closed intent classifier to open-world intent sets, which is crucial to task-oriented dialogue (TOD) systems. Previous methods address them by fine-tuning discriminative models. Recently, although some studies have been exploring the application of large language models (LLMs) represented by ChatGPT to various downstream tasks, it is still unclear for the ability of ChatGPT to discover and incrementally extent OOD intents. In this paper, we comprehensively evaluate ChatGPT on OOD intent discovery and GID, and then outline the strengths and weaknesses of ChatGPT. Overall, ChatGPT exhibits consistent advantages under zero-shot settings, but is still at a disadvantage compared to fine-tuned models. More deeply, through a series of analytical experiments, we summarize and discuss the challenges faced by LLMs including clustering, domain-specific understanding, and cross-domain in-context learning scenarios. Finally, we provide empirical guidance for future directions to address these challenges.

Yutao Zhu, Xingshuo Zhang, Maosen Zhang et al.

The advancement of large language models (LLMs) has significantly accelerated the development of search agents capable of autonomously gathering information through multi-turn web interactions. Various benchmarks have been proposed to evaluate such agents. However, existing benchmarks often construct queries backward from answers, producing unnatural tasks misaligned with real-world needs. Moreover, these benchmarks tend to focus on either locating specific information or aggregating information from multiple sources, while relying on static answer sets prone to data contamination. To bridge these gaps, we introduce GISA, a benchmark for General Information-Seeking Assistants comprising 373 human-crafted queries that reflect authentic information-seeking scenarios. GISA features four structured answer formats (item, set, list, and table), enabling deterministic evaluation. It integrates both deep reasoning and broad information aggregation within unified tasks, and includes a live subset with periodically updated answers to resist memorization. Notably, GISA provides complete human search trajectories for every query, offering gold-standard references for process-level supervision and imitation learning. Experiments on mainstream LLMs and commercial search products reveal that even the best-performing model achieves only 19.30\% exact match score, with performance notably degrading on tasks requiring complex planning and comprehensive information gathering. These findings highlight substantial room for future improvement.

Pei Wang, Keqing He, Yutao Mou et al.

Detecting out-of-domain (OOD) intents from user queries is essential for a task-oriented dialogue system. Previous OOD detection studies generally work on the assumption that plenty of labeled IND intents exist. In this paper, we focus on a more practical few-shot OOD setting where there are only a few labeled IND data and massive unlabeled mixed data that may belong to IND or OOD. The new scenario carries two key challenges: learning discriminative representations using limited IND data and leveraging unlabeled mixed data. Therefore, we propose an adaptive prototypical pseudo-labeling (APP) method for few-shot OOD detection, including a prototypical OOD detection framework (ProtoOOD) to facilitate low-resource OOD detection using limited IND data, and an adaptive pseudo-labeling method to produce high-quality pseudo OOD\&IND labels. Extensive experiments and analysis demonstrate the effectiveness of our method for few-shot OOD detection.

Xiaoshuai Song, Yutao Mou, Keqing He et al.

In a practical dialogue system, users may input out-of-domain (OOD) queries. The Generalized Intent Discovery (GID) task aims to discover OOD intents from OOD queries and extend them to the in-domain (IND) classifier. However, GID only considers one stage of OOD learning, and needs to utilize the data in all previous stages for joint training, which limits its wide application in reality. In this paper, we introduce a new task, Continual Generalized Intent Discovery (CGID), which aims to continuously and automatically discover OOD intents from dynamic OOD data streams and then incrementally add them to the classifier with almost no previous data, thus moving towards dynamic intent recognition in an open world. Next, we propose a method called Prototype-guided Learning with Replay and Distillation (PLRD) for CGID, which bootstraps new intent discovery through class prototypes and balances new and old intents through data replay and feature distillation. Finally, we conduct detailed experiments and analysis to verify the effectiveness of PLRD and understand the key challenges of CGID for future research.

Guanting Dong, Xiaoshuai Song, Yutao Zhu et al.

Following natural instructions is crucial for the effective application of Retrieval-Augmented Generation (RAG) systems. Despite recent advancements in Large Language Models (LLMs), research on assessing and improving instruction-following (IF) alignment within the RAG domain remains limited. To address this issue, we propose VIF-RAG, the first automated, scalable, and verifiable synthetic pipeline for instruction-following alignment in RAG systems. We start by manually crafting a minimal set of atomic instructions (<100) and developing combination rules to synthesize and verify complex instructions for a seed set. We then use supervised models for instruction rewriting while simultaneously generating code to automate the verification of instruction quality via a Python executor. Finally, we integrate these instructions with extensive RAG and general data samples, scaling up to a high-quality VIF-RAG-QA dataset (>100k) through automated processes. To further bridge the gap in instruction-following auto-evaluation for RAG systems, we introduce FollowRAG Benchmark, which includes approximately 3K test samples, covering 22 categories of general instruction constraints and four knowledge-intensive QA datasets. Due to its robust pipeline design, FollowRAG can seamlessly integrate with different RAG benchmarks. Using FollowRAG and eight widely-used IF and foundational abilities benchmarks for LLMs, we demonstrate that VIF-RAG markedly enhances LLM performance across a broad range of general instruction constraints while effectively leveraging its capabilities in RAG scenarios. Further analysis offers practical insights for achieving IF alignment in RAG systems. Our code and datasets are released at https://FollowRAG.github.io.

Xiaoshuai Song, Yanan Wu, Weixun Wang et al.

Self-Correction aims to enable large language models (LLMs) to self-verify and self-refine their initial responses without external feedback. However, LLMs often fail to effectively self-verify and generate correct feedback, further misleading refinement and leading to the failure of self-correction, especially in complex reasoning tasks. In this paper, we propose Program-driven Self-Correction (ProgCo). First, program-driven verification (ProgVe) achieves complex verification logic and extensive validation through self-generated, self-executing verification pseudo-programs. Then, program-driven refinement (ProgRe) receives feedback from ProgVe, conducts dual reflection and refinement on both responses and verification programs to mitigate misleading of incorrect feedback in complex reasoning tasks. Experiments on three instruction-following and mathematical benchmarks indicate that ProgCo achieves effective self-correction, and can be further enhance performance when combined with real program tools. We release our code at https://github.com/songxiaoshuai/progco.

Pei Wang, Yanan Wu, Zekun Wang et al.

Large Language Models (LLMs) have displayed massive improvements in reasoning and decision-making skills and can hold natural conversations with users. Recently, many tool-use benchmark datasets have been proposed. However, existing datasets have the following limitations: (1). Insufficient evaluation scenarios (e.g., only cover limited tool-use scenes). (2). Extensive evaluation costs (e.g., GPT API costs). To address these limitations, in this work, we propose a multi-granularity tool-use benchmark for large language models called MTU-Bench. For the "multi-granularity" property, our MTU-Bench covers five tool usage scenes (i.e., single-turn and single-tool, single-turn and multiple-tool, multiple-turn and single-tool, multiple-turn and multiple-tool, and out-of-distribution tasks). Besides, all evaluation metrics of our MTU-Bench are based on the prediction results and the ground truth without using any GPT or human evaluation metrics. Moreover, our MTU-Bench is collected by transforming existing high-quality datasets to simulate real-world tool usage scenarios, and we also propose an instruction dataset called MTU-Instruct data to enhance the tool-use abilities of existing LLMs. Comprehensive experimental results demonstrate the effectiveness of our MTU-Bench. Code and data will be released at https: //github.com/MTU-Bench-Team/MTU-Bench.git.

Xiaoshuai Song, Muxi Diao, Guanting Dong et al.

Large language models (LLMs) have demonstrated significant potential in advancing various fields of research and society. However, the current community of LLMs overly focuses on benchmarks for analyzing specific foundational skills (e.g. mathematics and code generation), neglecting an all-round evaluation of the computer science field. To bridge this gap, we introduce CS-Bench, the first multilingual (English, Chinese, French, German) benchmark dedicated to evaluating the performance of LLMs in computer science. CS-Bench comprises approximately 10K meticulously curated test samples, covering 26 subfields across 4 key areas of computer science, encompassing various task forms and divisions of knowledge and reasoning. Utilizing CS-Bench, we conduct a comprehensive evaluation of over 30 mainstream LLMs, revealing the relationship between CS performance and model scales. We also quantitatively analyze the reasons for failures in existing LLMs and highlight directions for improvements, including knowledge supplementation and CS-specific reasoning. Further cross-capability experiments show a high correlation between LLMs' capabilities in computer science and their abilities in mathematics and coding. Moreover, expert LLMs specialized in mathematics and coding also demonstrate strong performances in several CS subfields. Looking ahead, we envision CS-Bench serving as a cornerstone for LLM applications in the CS field and paving new avenues in assessing LLMs' diverse reasoning capabilities. The CS-Bench data and evaluation code are available at https://github.com/csbench/csbench.

Lixiong Qin, Mei Wang, Xuannan Liu et al.

With the comprehensive research conducted on various face analysis tasks, there is a growing interest among researchers to develop a unified approach to face perception. Existing methods mainly discuss unified representation and training, which lack task extensibility and application efficiency. To tackle this issue, we focus on the unified model structure, exploring a face generalist model. As an intuitive design, Naive Faceptor enables tasks with the same output shape and granularity to share the structural design of the standardized output head, achieving improved task extensibility. Furthermore, Faceptor is proposed to adopt a well-designed single-encoder dual-decoder architecture, allowing task-specific queries to represent new-coming semantics. This design enhances the unification of model structure while improving application efficiency in terms of storage overhead. Additionally, we introduce Layer-Attention into Faceptor, enabling the model to adaptively select features from optimal layers to perform the desired tasks. Through joint training on 13 face perception datasets, Faceptor achieves exceptional performance in facial landmark localization, face parsing, age estimation, expression recognition, binary attribute classification, and face recognition, achieving or surpassing specialized methods in most tasks. Our training framework can also be applied to auxiliary supervised learning, significantly improving performance in data-sparse tasks such as age estimation and expression recognition. The code and models will be made publicly available at https://github.com/lxq1000/Faceptor.

Pei Wang, Keqing He, Yejie Wang et al.

Out-of-domain (OOD) intent detection aims to examine whether the user's query falls outside the predefined domain of the system, which is crucial for the proper functioning of task-oriented dialogue (TOD) systems. Previous methods address it by fine-tuning discriminative models. Recently, some studies have been exploring the application of large language models (LLMs) represented by ChatGPT to various downstream tasks, but it is still unclear for their ability on OOD detection task.This paper conducts a comprehensive evaluation of LLMs under various experimental settings, and then outline the strengths and weaknesses of LLMs. We find that LLMs exhibit strong zero-shot and few-shot capabilities, but is still at a disadvantage compared to models fine-tuned with full resource. More deeply, through a series of additional analysis experiments, we discuss and summarize the challenges faced by LLMs and provide guidance for future work including injecting domain knowledge, strengthening knowledge transfer from IND(In-domain) to OOD, and understanding long instructions.

Lixiong Qin, Shilong Ou, Miaoxuan Zhang et al.

Faces and humans are crucial elements in social interaction and are widely included in everyday photos and videos. Therefore, a deep understanding of faces and humans will enable multi-modal assistants to achieve improved response quality and broadened application scope. Currently, the multi-modal assistant community lacks a comprehensive and scientific evaluation of face and human understanding abilities. In this paper, we first propose a hierarchical ability taxonomy that includes three levels of abilities. Then, based on this taxonomy, we collect images and annotations from publicly available datasets in the face and human community and build a semi-automatic data pipeline to produce problems for the new benchmark. Finally, the obtained Face-Human-Bench includes a development set and a test set, each with 1800 problems, supporting both English and Chinese. We conduct evaluations over 25 mainstream multi-modal large language models (MLLMs) with our Face-Human-Bench, focusing on the correlation between abilities, the impact of the relative position of targets on performance, and the impact of Chain of Thought (CoT) prompting on performance. We also explore which abilities of MLLMs need to be supplemented by specialist models. The dataset and evaluation code have been made publicly available at https://face-human-bench.github.io.

Xiaoshuai Song, Zhengyang Wang, Keqing He et al.

Knowledge editing (KE) aims to efficiently and precisely modify the behavior of large language models (LLMs) to update specific knowledge without negatively influencing other knowledge. Current research primarily focuses on white-box LLMs editing, overlooking an important scenario: black-box LLMs editing, where LLMs are accessed through interfaces and only textual output is available. In this paper, we first officially introduce KE on black-box LLMs and then propose a comprehensive evaluation framework to overcome the limitations of existing evaluations that are not applicable to black-box LLMs editing and lack comprehensiveness. To tackle privacy leaks of editing data and style over-editing in current methods, we introduce a novel postEdit framework, resolving privacy concerns through downstream post-processing and maintaining textual style consistency via fine-grained editing to original responses. Experiments and analysis on two benchmarks demonstrate that postEdit outperforms all baselines and achieves strong generalization, especially with huge improvements on style retention (average $+20.82\%\uparrow$).

Jinxu Zhao, Guanting Dong, Yueyan Qiu et al.

In a realistic dialogue system, the input information from users is often subject to various types of input perturbations, which affects the slot-filling task. Although rule-based data augmentation methods have achieved satisfactory results, they fail to exhibit the desired generalization when faced with unknown noise disturbances. In this study, we address the challenges posed by input perturbations in slot filling by proposing Noise-BERT, a unified Perturbation-Robust Framework with Noise Alignment Pre-training. Our framework incorporates two Noise Alignment Pre-training tasks: Slot Masked Prediction and Sentence Noisiness Discrimination, aiming to guide the pre-trained language model in capturing accurate slot information and noise distribution. During fine-tuning, we employ a contrastive learning loss to enhance the semantic representation of entities and labels. Additionally, we introduce an adversarial attack training strategy to improve the model's robustness. Experimental results demonstrate the superiority of our proposed approach over state-of-the-art models, and further analysis confirms its effectiveness and generalization ability.

YiFan Zhang, Shanglin Lei, Runqi Qiao et al.

The rapidly developing field of large multimodal models (LMMs) has led to the emergence of diverse models with remarkable capabilities. However, existing benchmarks fail to comprehensively, objectively and accurately evaluate whether LMMs align with the diverse needs of humans in real-world scenarios. To bridge this gap, we propose the Multi-Dimensional Insights (MDI) benchmark, which includes over 500 images covering six common scenarios of human life. Notably, the MDI-Benchmark offers two significant advantages over existing evaluations: (1) Each image is accompanied by two types of questions: simple questions to assess the model's understanding of the image, and complex questions to evaluate the model's ability to analyze and reason beyond basic content. (2) Recognizing that people of different age groups have varying needs and perspectives when faced with the same scenario, our benchmark stratifies questions into three age categories: young people, middle-aged people, and older people. This design allows for a detailed assessment of LMMs' capabilities in meeting the preferences and needs of different age groups. With MDI-Benchmark, the strong model like GPT-4o achieve 79% accuracy on age-related tasks, indicating that existing LMMs still have considerable room for improvement in addressing real-world applications. Looking ahead, we anticipate that the MDI-Benchmark will open new pathways for aligning real-world personalization in LMMs. The MDI-Benchmark data and evaluation code are available at https://mdi-benchmark.github.io/

Yutao Mou, Xiaoshuai Song, Keqing He et al.

Generalized intent discovery aims to extend a closed-set in-domain intent classifier to an open-world intent set including in-domain and out-of-domain intents. The key challenges lie in pseudo label disambiguation and representation learning. Previous methods suffer from a coupling of pseudo label disambiguation and representation learning, that is, the reliability of pseudo labels relies on representation learning, and representation learning is restricted by pseudo labels in turn. In this paper, we propose a decoupled prototype learning framework (DPL) to decouple pseudo label disambiguation and representation learning. Specifically, we firstly introduce prototypical contrastive representation learning (PCL) to get discriminative representations. And then we adopt a prototype-based label disambiguation method (PLD) to obtain pseudo labels. We theoretically prove that PCL and PLD work in a collaborative fashion and facilitate pseudo label disambiguation. Experiments and analysis on three benchmark datasets show the effectiveness of our method.