arXiv subjects

Xinlong Yang

Publications and source records attributed to Xinlong Yang.

12 recordsLinked to original sources

arXiv2026

TVIR: Building Deep Research Agents Towards Text-Visual Interleaved Report Generation

Deep Research Agents have shown strong capability in multi-step information retrieval, reasoning, and long-form report generation, but existing benchmarks and systems remain predominantly text-centric, with limited evaluation of whether visual elements are factually reliable and well aligned with the surrounding analysis. To address this gap, we introduce TVIR (Text-Visual Interleaved Report Generation), which includes TVIR-Bench, a benchmark of 100 expert-curated multimodal deep research tasks that require visual elements to serve specific analytical sub-goals, and TVIR-Agent, a hierarchical multi-agent framework that serves as a strong baseline for constructing outlines, retrieving images, generating charts with traceable sources, and composing reports through context-aware sequential writing. We further develop a dual-path evaluation framework that combines Textual Assessment and Visual Assessment. Experiments across nine deep research systems show that TVIR-Agent achieves strong overall performance, underscoring the importance of explicit multimodal design and evaluation for evidence-driven report generation.

cs.CL↗

arXiv2026

TriSpec: Ternary Speculative Decoding via Lightweight Proxy Verification

Inference efficiency in Large Language Models (LLMs) is fundamentally limited by their serial, autoregressive generation, especially as reasoning becomes a key capability and response sequences grow longer. Speculative decoding (SD) offers a powerful solution, providing significant speed-ups through its lightweight drafting and parallel verification mechanism. While existing work has nearly saturated improvements in draft effectiveness and efficiency, this paper advances SD from a new yet critical perspective: the verification cost. We propose TriSpec, a novel ternary SD framework that, at its core, introduces a lightweight proxy to significantly reduce computational cost by approving easily verifiable draft sequences and engaging the full target model only when encountering uncertain tokens. TriSpec can be integrated with state-of-the-art SD methods like EAGLE-3 to further reduce verification costs, achieving greater acceleration. Extensive experiments on the Qwen3 and DeepSeek-R1-Distill-Qwen/LLaMA families show that TriSpec achieves up to 35\% speedup over standard SD, with up to 50\% fewer target model invocations while maintaining comparable accuracy.

cs.LG↗

arXiv2026

O-Researcher: An Open Ended Deep Research Model via Multi-Agent Distillation and Agentic RL

The performance gap between closed-source and open-source large language models (LLMs) is largely attributed to disparities in access to high-quality training data. To bridge this gap, we introduce a novel framework for the automated synthesis of sophisticated, research-grade instructional data. Our approach centers on a multi-agent workflow where collaborative AI agents simulate complex tool-integrated reasoning to generate diverse and high-fidelity data end-to-end. Leveraging this synthesized data, we develop a two-stage training strategy that integrates supervised fine-tuning with a novel reinforcement learning method, designed to maximize model alignment and capability. Extensive experiments demonstrate that our framework empowers open-source models across multiple scales, enabling them to achieve new state-of-the-art performance on the major deep research benchmark. This work provides a scalable and effective pathway for advancing open-source LLMs without relying on proprietary data or models.

cs.CL↗

arXiv2025

Position: Reasoning After Perception Means Reasoning Without Vision

A common belief in multimodal research is that the perceptual weaknesses of vision--language models can be compensated by stronger language reasoning (e.g., chain-of-thought, in-context learning, or external tools). We challenge this assumption. We argue that for a broad class of visual tasks hard to specify in language, failures stem from a structural fatality where the temporal decision of \textit{when} to reason strictly dictates the spatial constraint of \textit{where} reasoning takes place. When visual reasoning is deferred to language generation, current architectures do not merely delay computation; they displace it from the continuous visual representation to a discrete textual space. Consequently, the sequential ``Perception-then-Reasoning'' paradigm degenerates perception into a passive, one-off feature encoding process, rendering it functionally equivalent to ``Reasoning-in-Text-Space'', where task-critical spatial signals are collapsed before reasoning begins. We substantiate this claim with the Turing Eye Test (TET): tasks that must be resolved in \emph{visual space} and are hard to verbalize; results show text-only reasoning cannot remedy these perceptual failures. Our findings suggest rethinking the architectural divide: shifting from reasoning \textit{about} perception to reasoning \textit{within} perception. This facilitates actively reasoning-driven perception that operates directly on pixel-level visual representations, rather than within a collapsed textual space.

cs.CV↗

arXiv2025

Comparison Research of Millimeter-Wave/Infrared Co-aperture Reflector Antenna Systems Based on a Specialized Film

This paper presents a novel co-aperture reflector antenna operating in millimeter-wave (MMW) and infrared (IR) for cloud detection radar. The proposed design combines a back-fed dual-reflector antenna, an IR optical reflection system, and a specialize thin film with IR-reflective/MMW-transmissive properties. Simulations demonstrate a gain exceeding 50 dBi within 94 GHz plush or minus 500 MHz bandwidth, with 0.46{\deg} beamwidth in both azimuth (E-plane) and elevation (H-plane) and sidelobe levels below -25 dB. This co-aperture architecture addresses the limitations of standalone MMW and IR radars, enabling high-resolution cloud microphysical parameter retrieval while minimizing system footprint. The design lays a foundation for airborne/spaceborne multi-mode detection systems.

eess.SP↗

arXiv2025

Qwen2.5-1M Technical Report

We introduce Qwen2.5-1M, a series of models that extend the context length to 1 million tokens. Compared to the previous 128K version, the Qwen2.5-1M series have significantly enhanced long-context capabilities through long-context pre-training and post-training. Key techniques such as long data synthesis, progressive pre-training, and multi-stage supervised fine-tuning are employed to effectively enhance long-context performance while reducing training costs. To promote the use of long-context models among a broader user base, we present and open-source our inference framework. This framework includes a length extrapolation method that can expand the model context lengths by at least four times, or even more, without additional training. To reduce inference costs, we implement a sparse attention method along with chunked prefill optimization for deployment scenarios and a sparsity refinement method to improve precision. Additionally, we detail our optimizations in the inference engine, including kernel optimization, pipeline parallelism, and scheduling optimization, which significantly enhance overall inference performance. By leveraging our inference framework, the Qwen2.5-1M models achieve a remarkable 3x to 7x prefill speedup in scenarios with 1 million tokens of context. This framework provides an efficient and powerful solution for developing applications that require long-context processing using open-source models. The Qwen2.5-1M series currently includes the open-source models Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M, as well as the API-accessed model Qwen2.5-Turbo. Evaluations show that Qwen2.5-1M models have been greatly improved in long-context tasks without compromising performance in short-context scenarios. Specifically, the Qwen2.5-14B-Instruct-1M model significantly outperforms GPT-4o-mini in long-context tasks and supports contexts eight times longer.

cs.CL↗

arXiv2024

Amphista: Bi-directional Multi-head Decoding for Accelerating LLM Inference

Large Language Models (LLMs) inherently use autoregressive decoding, which lacks parallelism in inference and results in significantly slow inference speed. While methods such as Medusa constructs parallelized heads, they lack adequate information interaction across different prediction positions. To overcome this limitation, we introduce Amphista, an enhanced speculative decoding framework that builds upon Medusa. Specifically, Amphista models an Auto-embedding Block capable of parallel inference, incorporating bi-directional attention to enable interaction between different drafting heads. Additionally, Amphista integrates Staged Adaptation Layers, which ensure a seamless transition of semantic information from the target model's autoregressive inference to the drafting heads' non-autoregressive inference, effectively achieving paradigm shift and feature fusion. Experimental results on Vicuna models using MT-Bench and Spec-Bench demonstrate that Amphista achieves substantial acceleration while maintaining generation quality. On MT-Bench, Amphista delivers up to 2.75$\times$ speedup over vanilla autoregressive decoding and 1.40$\times$ over Medusa on Vicuna 33B in wall-clock time.

cs.AI↗

arXiv2024

Magnetic ground state of monolayer CeI$_{2}$: occupation matrix control and DFT+U calculations

The magnetic ground state is crucial for the applications of the two-dimension magnets as it decides fundamental magnetic properties of the material, such as magnetic order, magnetic transition temperature, and low-energy excitation of the spin waves. However, the simulations for magnetism of local-electron systems are challenging due to the existence of metastable states. In this study, occupation matrix control (OMC) and density functional theory plus Hubbard $U$ calculations are applied to investigate the magnetic ground state of monolayer CeI$_{2}$. Following the predicted ferromagnetic (FM) order, the FM ground state and the FM metastable states are identified and found to have different values of the magnetic parameters. Based on the calculated magnetic parameters of the FM ground state, the Curie temperature is estimated to be $128$ K for monolayer CeI$_{2}$. When spin-orbit coupling (SOC) is considered, the FM ground state is further confirmed to contain both off-plane and in-plane components of magnetization. SOC is shown to be essential for reasonably describing not only magnetic anisotropy but also local electronic orbital state of monolayer CeI$_{2}$.

cond-mat.mtrl-sci↗

arXiv2024

Monolayer Fe$_{3}$GaX$_{2}$ (X=I, Br, Sb): high-temperature two-dimensional magnets and a novel partially ordered spin state

We systematically investigated the effects of charge doping and strain on monolayer Fe$_3$GaTe$_2$, and proposed three new novel two-dimensional magnetic materials: monolayer Fe$_3$GaX$_2$ (X=I, Br, Sb). We found that both strain and charge doping can tune the magnetic interactions, and the tuning by charge doping is more significant. Differential charge analysis revealed that the doped charges predominantly accumulate around Te atoms. Based on this insight, we introduced Fe$_{3}$GaI$_{2}$, Fe$_{3}$GaBr$_{2}$, and Fe$_{3}$GaSb$_{2}$ monolayers. The Fe$_{3}$GaI$_{2}$ and Fe$_{3}$GaBr$_{2}$ monolayers contain I and Br atoms rather than Te atoms, emulate electron-doped Fe$_{3}$GaTe$_{2}$ monolayer, resulting in notably high T$_c$ values of 867 K and 844 K, respectively. In contrast, the Fe$_3$GaSb$_2$ monolayer mimics hole-doped Fe$_{3}$GaTe$_{2}$ monolayer, presents a mix of FM and antiferromagnetic interactions, manifesting a distinctive partially ordered magnetic state. Our study demonstrates that substitution atoms based on the charge-doping effect offer a promising approach for predicting new magnetic materials. The proposed Fe$_{3}$GaI$_{2}$, Fe$_{3}$GaBr$_{2}$, and Fe$_{3}$GaSb$_{2}$ monolayers hold great potential for spintronics applications, and may stimulate the pursuit of new types of spin liquid.

cond-mat.mtrl-sci↗

arXiv2024

Interlayer magnetic interactions and ferroelectricity in $\pi$/3-twisted CrX$_2$ (X = Se, Te) bilayers

Recently, two-dimensional (2D) bilayer magnetic systems have been widely studied. Their interlayer magnetic interactions play a vital role in the magnetic properties. In this paper, we theoretically studied the interlayer magnetic interactions, magnetic states and ferroelectricity of $\pi$/3-twisted CrX$_2$ (X = Se, Te) bilayers ($\pi$/3-CrX$_2$). Our study reveals that the lateral shift could switch the magnetic state of the $\pi$/3-CrSe$_2$ between interlayer ferromagnetic and antiferromagnetic, while just tuning the strength of the interlayer antiferromagnetic interactions in $\pi$/3-CrTe$_2$. Furthermore, the lateral shift can alter the off-plane electric polarization in both $\pi$/3-CrSe$_2$ and $\pi$/3-CrTe$_2$. These results show that stacking is an effective way to tune both the magnetic and ferroelectric properties of 1T-CrX$_2$ bilayers, making the 1T-CrX$_2$ bilayers hold promise for 2D spintronic devices.

cond-mat.mtrl-sci↗

arXiv2023

When Parameter-efficient Tuning Meets General-purpose Vision-language Models

Instruction tuning has shown promising potential for developing general-purpose AI capabilities by using large-scale pre-trained models and boosts growing research to integrate multimodal information for creative applications. However, existing works still face two main limitations: the high training costs and heavy computing resource dependence of full model fine-tuning, and the lack of semantic information in instructions, which hinders multimodal alignment. Addressing these challenges, this paper proposes a novel approach to utilize Parameter-Efficient Tuning for generAl-purpose vision-Language models, namely PETAL. PETAL revolutionizes the training process by requiring only 0.5% of the total parameters, achieved through a unique mode approximation technique, which significantly reduces the training costs and reliance on heavy computing resources. Furthermore, PETAL enhances the semantic depth of instructions in two innovative ways: 1) by introducing adaptive instruction mixture-of-experts(MOEs), and 2) by fortifying the score-based linkage between parameter-efficient tuning and mutual information. Our extensive experiments across five multimodal downstream benchmarks reveal that PETAL not only outperforms current state-of-the-art methods in most scenarios but also surpasses full fine-tuning models in effectiveness. Additionally, our approach demonstrates remarkable advantages in few-shot settings, backed by comprehensive visualization analyses. Our source code is available at: https://github. com/melonking32/PETAL.

cs.CL↗

arXiv2023

Parameter-efficient Tuning of Large-scale Multimodal Foundation Model

Driven by the progress of large-scale pre-training, parameter-efficient transfer learning has gained immense popularity across different subfields of Artificial Intelligence. The core is to adapt the model to downstream tasks with only a small set of parameters. Recently, researchers have leveraged such proven techniques in multimodal tasks and achieve promising results. However, two critical issues remain unresolved: how to further reduce the complexity with lightweight design and how to boost alignment between modalities under extremely low parameters. In this paper, we propose A graceful prompt framework for cross-modal transfer (Aurora) to overcome these challenges. Considering the redundancy in existing architectures, we first utilize the mode approximation to generate 0.1M trainable parameters to implement the multimodal prompt tuning, which explores the low intrinsic dimension with only 0.04% parameters of the pre-trained model. Then, for better modality alignment, we propose the Informative Context Enhancement and Gated Query Transformation module under extremely few parameters scenes. A thorough evaluation on six cross-modal benchmarks shows that it not only outperforms the state-of-the-art but even outperforms the full fine-tuning approach. Our code is available at: https://github.com/WillDreamer/Aurora.

cs.CV↗