Bingguang Hao

Ji Zhang, Lianli Gao, Bingguang Hao et al.

Out-of-distribution (OOD) detection aims to detect "unknown" data whose labels have not been seen during the in-distribution (ID) training process. Recent progress in representation learning gives rise to distance-based OOD detection that recognizes inputs as ID/OOD according to their relative distances to the training data of ID classes. Previous approaches calculate pairwise distances relying only on global image representations, which can be sub-optimal as the inevitable background clutter and intra-class variation may drive image-level representations from the same ID class far apart in a given representation space. In this work, we overcome this challenge by proposing Multi-scale OOD DEtection (MODE), a first framework leveraging both global visual information and local region details of images to maximally benefit OOD detection. Specifically, we first find that existing models pretrained by off-the-shelf cross-entropy or contrastive losses are incompetent to capture valuable local representations for MODE, due to the scale-discrepancy between the ID training and OOD detection processes. To mitigate this issue and encourage locally discriminative representations in ID training, we propose Attention-based Local PropAgation (ALPA), a trainable objective that exploits a cross-attention mechanism to align and highlight the local regions of the target objects for pairwise examples. During test-time OOD detection, a Cross-Scale Decision (CSD) function is further devised on the most discriminative multi-scale representations to distinguish ID/OOD data more faithfully. We demonstrate the effectiveness and flexibility of MODE on several benchmarks -- on average, MODE outperforms the previous state-of-the-art by up to 19.24% in FPR, 2.77% in AUROC. Code is available at https://github.com/JimZAI/MODE-OOD.

Haoyue Bai, Dong Wang, Long Chen et al.

Large language model-based web agents have demonstrated strong performance on realistic web interaction tasks. However, existing evaluations are predominantly conducted under relatively stable and well-behaved interaction conditions, which may overestimate agent robustness. High task success in such idealized settings does not necessarily reflect performance under realistic web interaction. To address this limitation, we introduce a diagnostic stress-testing benchmark for web agents. We first construct realistic and controllable web environments that provide clean and stable interaction workflows as reference baselines. We then introduce structured and controlled perturbations that emulate interaction variability, including shifting layouts, altered interaction semantics, and execution disruptions. By comparing agent behavior between clean and perturbed settings, our framework enables systematic diagnosis of robustness under what-if interaction scenarios. Through extensive evaluation of state-of-the-art multimodal web agents, we show that stress-based evaluation exposes failure modes and substantial robustness gaps that remain hidden under clean benchmark conditions.

Bingguang Hao, Zengzhuang Xu, Yuntao Wen et al.

The advancement of LLM agents with tool-use capabilities requires diverse and complex training corpora. Existing data generation methods, which predominantly follow a paradigm of random sampling and shallow generation, often yield simple and homogeneous trajectories that fail to capture complex, implicit logical dependencies. To bridge this gap, we introduce HardGen, an automatic agentic pipeline designed to generate hard tool-use training samples with verifiable reasoning. Firstly, HardGen establishes a dynamic API Graph built upon agent failure cases, from which it samples to synthesize hard traces. Secondly, these traces serve as conditional priors to guide the instantiation of modular, abstract advanced tools, which are subsequently leveraged to formulate hard queries. Finally, the advanced tools and hard queries enable the generation of verifiable complex Chain-of-Thought (CoT), with a closed-loop evaluation feedback steering the continuous refinement of the process. Extensive evaluations demonstrate that a 4B parameter model trained with our curated dataset achieves superior performance compared to several leading open-source and closed-source competitors (e.g., GPT-5.2, Gemini-3-Pro and Claude-Opus-4.5). Our code, models, and dataset will be open-sourced to facilitate future research.

Zengzhuang Xu, Bingguang Hao, Zechuan Wang et al.

Function calling (FC) empowers large language models (LLMs) and autonomous agents to interface with external tools, a critical capability for solving complex, real-world problems. As this ability becomes increasingly central to advanced AI systems, the need for high-quality, multi-turn training data to develop and refine it cannot be overstated. Existing data synthesis methods, such as random environment sampling or multi-agent role-playing, are not powerful enough to generate high-quality data in real-world environments. Practical challenges come in three folds: targeted data synthesis, hard query construction, and multi-turn logical dependency. To address these structural deficiencies, we present FunReason-MT, a novel data synthesis framework for real-world multi-turn tool use. FunReason-MT resolves the complexity barrier in multi-turn FC data by employing 1) Environment-API Graph Interactions to gather varied high-quality trajectories with targeted tool, 2) Advanced Tool-Query Synthesis to simplify hard query construction, and 3) Guided Iterative Chain for sophisticated CoT generation. Evaluations on Berkeley Function-Calling Leaderboard (BFCLv3) demonstrate the power of our framework: a 4B model built upon FunReason-MT generated data achieves state-of-the-art performance among comparable-sized models. Further performance improvements on BFCLv4 confirm that FunReason-MT provides a reliable and robust source for agentic learning.

Bingguang Hao, Zengzhuang Xu, Maolin Wang et al.

The effective training of Large Language Models (LLMs) for function calling faces a critical challenge: balancing exploration of complex reasoning paths with stable policy optimization. Standard methods like Supervised Fine-Tuning (SFT) fail to instill robust reasoning, and traditional Reinforcement Learning (RL) struggles with inefficient exploration. We propose \textbf{EGPO}, a new RL framework built upon Group Relative Policy Optimization (GRPO), designed to address this challenge directly. The core of EGPO is an entropy-enhanced advantage function that integrates the entropy of the model's Chain-of-Thought (CoT) into the policy gradient computation. This encourages the generation of diverse reasoning strategies. To maintain optimization direction, the entropy bonus is carefully constrained by a clipping mechanism. Complemented by a strict, binary reward signal, EGPO effectively guides the model towards discovering structured and accurate tool invocation patterns. On the challenging Berkeley Function Calling Leaderboard (BFCL), a 4B-parameter model trained with EGPO sets a new state-of-the-art among models of comparable size, surpassing a range of strong competitors, including GPT-4o and Gemini-2.5.

Zhenyu Zhang, Bingguang Hao, Jinpeng Li et al.

Most large language models (LLMs) are sensitive to prompts, and another synonymous expression or a typo may lead to unexpected results for the model. Composing an optimal prompt for a specific demand lacks theoretical support and relies entirely on human experimentation, which poses a considerable obstacle to popularizing generative artificial intelligence. However, there is no systematic analysis of the stability of LLMs in resisting prompt perturbations in real-world scenarios. In this work, we propose to evaluate the ease-of-use of LLMs and construct E-Bench, simulating the actual situation of human use from synonymous perturbation (including paraphrasing, simplification, and colloquialism) and typographical perturbation (such as typing). On this basis, we also discuss the combination of these two types of perturbation and analyze the main reasons for performance degradation. Experimental results indicate that with the increase of model size, although the ease-of-use are significantly improved, there is still a long way to go to build a sufficiently user-friendly model.