arXiv subjects

Xin Jiang

Publications and source records attributed to Xin Jiang.

At least 37 records · Page 2Linked to original sources

arXiv2026

YouZhi: Towards High-Concurrency Financial LLMs via Adaptive GQA-to-MLA Transition

Large language models (LLMs) drive significant financial innovations, yet their high-concurrency deployment is severely bottlenecked by KV cache memory overhead, which inflates infrastructure costs and throttles scalability. To address this, we propose YouZhi-LLM, a highly efficient financial LLM empowered by a comprehensive structural transition and training pipeline natively built on the Huawei Ascend ecosystem. At its algorithmic core, YouZhi-LLM features a layer-adaptive GQA-to-MLA transition framework that dynamically assigns per-layer FreqFold sizes, maximizing KV-cache compression while minimizing perplexity degradation. To recover representation capacity and inject domain expertise, the Ascend-based training pipeline seamlessly integrates generalized knowledge distillation with financial-specific supervised fine-tuning. Evaluations demonstrate the superiority of this systematic approach, with the adaptive transition reducing perplexity degradation by up to 35% over uniform baselines. Crucially, when evaluated on Ascend NPUs via vLLM-Ascend, the massive KV-cache reduction translates directly into deployment efficiency. Compared to their respective base models, YouZhi-7B yields a 12.3% improvement in average financial benchmark score alongside a 2.69$\times$ increase in maximum concurrency; similarly, YouZhi-14B achieves a 7.0% accuracy gain and a 2.43$\times$ concurrency boost, establishing a new paradigm for cost-effective, high-throughput financial inference.

cs.CL↗

arXiv2026

Exact Bulk-Boundary Pairs in AdS/CFT

We show that for a CFT$_D$ on a flat open solid torus, the two point function in the Weyl frame is exactly paired with a finite geodesic lying entirely in the AdS$_{D+1}$ bulk interior. This relation is exact and requires neither large $N$, strong coupling, nor heavy operators. The exactness is that of conformal kinematics; no semiclassical bulk dynamics is assumed. The standard boundary-anchored relation is a singular limit of the exact pair. For the free scalar, a mode expansion along $S^1$ generates an infinite tower of effective masses on $H_{D-1}$, whose intricate propagators resum exactly to the same simple higher-dimensional geodesic expression. Together with another exact pair between disjoint entanglement entropy and entanglement wedge cross-section found on the same open solid torus, this result points toward a broader exact-pair program in AdS/CFT.

hep-th↗

arXiv2026

Timelike entanglement entropy Revisited

We present an operator-algebraic definition for timelike entanglement entropy in QFT under a few mild postulates. This rigorously defined timelike entanglement entropy is real-valued due to the timelike tube theorem. We further demonstrate why the timelike entanglement entropy should be real-valued from both path integral argument and holography perspective.

hep-th↗

arXiv2026

Local Linear Convergence of the Alternating Direction Method of Multipliers for Semidefinite Programming under Strict Complementarity

We investigate the local linear convergence properties of the Alternating Direction Method of Multipliers (ADMM) when applied to Semidefinite Programming (SDP). A longstanding belief suggests that ADMM is only capable of solving SDPs to moderate accuracy, primarily due to its sublinear worst-case complexity and empirical observations of slow convergence. We challenge this notion by introducing a new sufficient condition for local linear convergence: as long as the converged primal-dual optimal solutions satisfy strict complementarity, ADMM attains local linear convergence, independent of nondegeneracy conditions. Our proof is based on a direct local linearization of the ADMM operator and a refined error bound for the projection onto the positive semidefinite cone, improving previous bounds and revealing the anisotropic nature of projection residuals. Extensive numerical experiments confirm the significance of our theoretical results, demonstrating that ADMM achieves local linear convergence and computes high-accuracy solutions in a variety of SDP instances, including those cases where nondegeneracy fails. Furthermore, we identify where ADMM struggles, linking these difficulties with near violations of strict complementarity-a phenomenon that parallels recent findings in linear programming.

math.OC↗

arXiv2026

Mutual Enhancement Between Global Tokens and Patch Tokens: From Theory to Practice

Accurate and effective discrete image tokenization is crucial for long image sequence processing. However, current methods rigidly compress all content at a fixed rate, ignoring the variable information density of images and leading to either redundancy or information loss. Inspired by information entropy, we propose TaTok, a Theoretically grounded adaptive image Tokenization framework. We rigorously identify two key drawbacks in existing methods: information insufficiency when reconstructing images with patch tokens alone, and information redundancy among patch tokens. To address these, we introduce global tokens that model mutual information across patch tokens, and a Dynamic Token Filtering (DTF) algorithm based on cumulative conditional entropy to eliminate redundancy. Experiments confirm TaTok's state-of-the-art performance, delivering a 1.3x gFID improvement and 8.7x inference speedup. By allocating tokens according to information richness, TaTok enables more compressed yet accurate image tokenization, offering valuable insights for future research.

cs.CV↗

arXiv2026

Predictive and feedback signals differently shape the formation of group-level and individualized language representations

Adults vary greatly in how effectively they learn a new language, but the signals driving the learning processes and individual differences remain unclear. Over seven days, we tracked behavioral learning and collected fMRI data from 102 adults as they learned an artificial language with corrective feedback. We trained matched transformer models with prediction, feedback, or combined objectives and compared their internal representations to brain activity. Representations derived from the prediction-focused model accounted for the largest share of unique neural variance at the group level, despite the human task being feedback-based. Throughout model training, both objectives showed a shift in brain-model alignment from sensory to higher-order language and associative networks, indicating abstraction processing. Conversely, neural patterns related to the feedback model were most useful for predicting individual generalization outcomes on Day 7. These findings support a multi-signal model of adult language learning, in which prediction shapes a common neural learning architecture across learners, whereas feedback-related mechanisms better explain individual differences over time.

q-bio.NC↗

arXiv2026

Enhancement of vacuum-ultraviolet dispersive-wave emission using gas-filled tapered hollow-core fibers

The recent breakthroughs in laser-driving 229Th nuclear transition have created an urgent demand for coherent vacuum-ultraviolet (VUV) sources delivering high spectral brightness at the critical 148.38 nm isomer energy. However, generating sufficient photon flux to overcome the low nuclear excitation probability remains a challenge for compact setups. While resonant dispersive wave emission in gas-filled hollow-core fibers offers a promising route, standard capillaries face a fundamental trade-off: maximizing input coupling requires large core diameters, whereas efficient nonlinear VUV conversion demands the high intensities using small cores. Here, we resolve this conflict using a gas-filled tapered capillary fiber. This architecture utilizes a longitudinally decreasing core diameter to combine a large input aperture with adiabatic field concentration, thereby continuously enhancing the nonlinear interaction. Experimentally, we demonstrate a widely tunable source (135-240 nm) that achieves a twofold efficiency enhancement specifically at the 148.38 nm wavelength compared to uniform geometries. By providing a scalable route to high-flux VUV generation, this work establishes a critical tabletop tool for advancing solid-state nuclear clocks and time-resolved spectroscopy.

physics.optics↗

arXiv2026

Gradually Excavating External Knowledge for Implicit Complex Question Answering

Recently, large language models (LLMs) have gained much attention for the emergence of human-comparable capabilities and huge potential. However, for open-domain implicit question-answering problems, LLMs may not be the ultimate solution due to the reasons of: 1) uncovered or out-of-date domain knowledge, 2) one-shot generation and hence restricted comprehensiveness. To this end, this work proposes a gradual knowledge excavation framework for open-domain complex question answering, where LLMs iteratively and actively acquire external information, and then reason based on acquired historical knowledge. Specifically, during each step of the solving process, the model selects an action to execute, such as querying external knowledge or performing a single logical reasoning step, to gradually progress toward a final answer. Our method can effectively leverage plug-and-play external knowledge and dynamically adjust the strategy for solving complex questions. Evaluated on the StrategyQA dataset, our method achieves 78.17% accuracy with less than 6% parameters of its competitors, setting new SOTA for ~10B-scale LLMs.

cs.CL↗

arXiv2026

Deep-ultraviolet Cherenkov radiation in all-normal-dispersion waveguide enabled by spatial-temporal dynamics

Nonlinear propagation of ultrashort pulses in multi-mode waveguides, featuring complex spatial-temporal dynamics, provides new degrees of freedom in the fields of nonlinear optics and ultrafast lasers. Here, we demonstrate a new scheme of ultraviolet Cherenkov (dispersive-wave) radiation in a gas-filled capillary with unprecedently-high pulse energy, enabled by spatial-temporal dynamics. We found that mJ-level, 40-fs pulses, launched into a large-core capillary filled with high-pressure noble gas, would experience self-phase-modulation and self-steepening effects in this normal-dispersion waveguide, leading to high-intensity shock wave generation and asymmetric spectral broadening. Spatial-temporal dynamics, stemming from strong nonlinear inter-mode coupling, causes spatial shrink and temporal deceleration of the pulse which dramatically alter the capillary dispersion landscape. As a result, a phase-matching point can be created in the ultraviolet, giving rise to the radiation of multi-mode dispersive waves with 100-μJ-level pulse energies and few-fs pulse widths. Our findings inspire new insights into multi-mode nonlinear optics, and the demonstrated high-energy ultraviolet light source with broadband tunability and compact set-up configuration, may find a few applications in time-resolved spectroscopy, ultrafast electronics and femtosecond chemistry.

physics.optics↗

arXiv2026

VLM-Pruner: Buffering for Spatial Sparsity in an Efficient VLM Centrifugal Token Pruning Paradigm

Vision-language models (VLMs) excel at image understanding tasks, but the large number of visual tokens imposes significant computational costs, hindering deployment on mobile devices. Many pruning methods rely solely on token importance and thus overlook inter-token redundancy, retaining numerous duplicated tokens and wasting capacity. Although some redundancy-aware approaches have been proposed, they often ignore the spatial relationships among visual tokens. This can lead to overly sparse selections of retained tokens that fail to adequately cover the regions of target objects. To address these limitations, we propose VLM-Pruner, a training-free token pruning algorithm that explicitly balances redundancy and spatial sparsity. We introduce a centrifugal token pruning paradigm that enables near-to-far selection while prioritizing the preservation of fine-grained object details. Moreover, we design a Buffering for Spatial Sparsity (BSS) criterion that defers the selection of spatially distant tokens. We further adopt a parallel greedy strategy to conduct token selection efficiently. To mitigate information loss from pruning, we selectively fuse salient information from the discarded tokens into the retained ones. Comprehensive comparisons demonstrate that VLM-Pruner consistently outperforms strong baselines across five VLMs with an 88.9\% pruning rate, while delivering an end-to-end inference speedup. The code is available at https://github.com/Casey-bit/VLMPruner.

cs.CV↗

arXiv2026

A Survey on the Optimization of Large Language Model-based Agents

With the rapid development of Large Language Models (LLMs), LLM-based agents have been widely adopted in various fields, becoming essential for autonomous decision-making and interactive tasks. However, current work typically relies on prompt design or fine-tuning strategies applied to vanilla LLMs, which often leads to limited effectiveness or suboptimal performance in complex agent-related environments. Although LLM optimization techniques can improve model performance across many general tasks, they lack specialized optimization towards critical agent functionalities such as long-term planning, dynamic environmental interaction, and complex decision-making. Although numerous recent studies have explored various strategies to optimize LLM-based agents for complex agent tasks, a systematic review summarizing and comparing these methods from a holistic perspective is still lacking. In this survey, we provide a comprehensive review of LLM-based agent optimization approaches, categorizing them into parameter-driven and parameter-free methods. We first focus on parameter-driven optimization, covering fine-tuning-based optimization, reinforcement learning-based optimization, and hybrid strategies, analyzing key aspects such as trajectory data construction, fine-tuning techniques, reward function design, and optimization algorithms. Additionally, we briefly discuss parameter-free strategies that optimize agent behavior through prompt engineering and external knowledge retrieval. Finally, we summarize the datasets and benchmarks used for evaluation and tuning, review key applications of LLM-based agents, and discuss major challenges and promising future directions. Our repository for related references is available at https://github.com/YoungDubbyDu/LLM-Agent-Optimization.

cs.AI↗

arXiv2026

ToolACE-MT: Non-Autoregressive Generation for Agentic Multi-Turn Interaction

Agentic task-solving with Large Language Models (LLMs) requires multi-turn, multi-step interactions, often involving complex function calls and dynamic user-agent exchanges. Existing simulation-based data generation methods for such scenarios rely heavily on costly autoregressive interactions between multiple LLM agents, thereby compromising the practical efficiency of agentic data generation. In this paper, we propose ToolACE-MT, a novel Non-Autoregressive Iterative Generation framework for constructing high-quality multi-turn agentic dialogues. ToolACE-MT generates full conversational trajectories through three stages: coarse-grained initialization, iterative refinement, and offline verification. The initialization phase builds a structurally complete yet semantically coarse dialogue skeleton; the iterative refinement phase introduces realistic complexities and continued refinement via mask-and-fill operations; and the offline verification phase ensures correctness and coherence via rule- and model-based checks. Experiments demonstrate that ToolACE-MT enables efficient, effective and generalizable agentic data generation, offering a new paradigm for high-quality data construction in tool-augmented LLM scenarios.

cs.CL↗

arXiv2026

FLM-Audio: Natural Monologues Improves Native Full-Duplex Chatbots via Dual Training

Full-duplex dialog models aim to listen and speak simultaneously, delivering rapid responses to dynamic user input. Among different solutions to full-duplexity, a native solution merges multiple channels in each time step, achieving the lowest latency. However, prevailing designs break down the textual monologue sentences for word-level alignment with audio streams, which degrades language modeling abilities. To help address this issue, we introduce "contiguous monologues", which are composed by continuous sentences and "waiting" intervals, mimicking human-like cognitive behavior in dialogs. We find a proper training paradigm to be critical for semantically aligning contiguous monologues with audio. To this end, we develop a "dual" training paradigm that alternates the position of the monologues, either leading or trailing the audio, across different training stages. A combination of our contiguous monologue and dual training strategy is applied in developing FLM-Audio, our 7B spoken dialog chatbot with native full-duplexity. As confirmed by experimental results, FLM-Audio achieves superior response qualities and chatting experiences while requiring significantly less training data.

cs.SD↗

arXiv2026

EgoMem: Lifelong Memory Agent for Full-duplex Omnimodal Models

We introduce EgoMem, the first lifelong memory agent tailored for full-duplex models that process real-time omnimodal streams. EgoMem enables real-time models to recognize multiple users directly from raw audiovisual streams, to provide personalized response, and to maintain long-term knowledge of users' facts, preferences, and social relationships extracted from audiovisual history. EgoMem operates with three asynchronous processes: (i) a retrieval process that dynamically identifies user via face and voice, and gathers relevant context from a long-term memory; (ii) an omnimodal dialog process that generates personalized audio responses based on the retrieved context; and (iii) a memory management process that automatically detects dialog boundaries from omnimodal streams, and extracts necessary information to update the long-term memory. Unlike existing memory agents for LLMs, EgoMem relies entirely on raw audiovisual streams, making it especially suitable for lifelong, real-time, and embodied scenarios. Experimental results demonstrate that EgoMem's retrieval and memory management modules achieve over 95% accuracy on the test set. When integrated with a fine-tuned RoboEgo omnimodal chatbot, the system achieves fact-consistency scores above 87% in real-time personalized dialogs, establishing a strong baseline for future research.

cs.AI↗

arXiv2026

DreamVAR: Taming Reinforced Visual Autoregressive Model for High-Fidelity Subject-Driven Image Generation

Recent advances in subject-driven image generation using diffusion models have attracted considerable attention for their remarkable capabilities in producing high-quality images. Nevertheless, the potential of Visual Autoregressive (VAR) models, despite their unified architecture and efficient inference, remains underexplored. In this work, we present DreamVAR, a novel framework for subject-driven image synthesis built upon a VAR model that employs next-scale prediction. Technically, multi-scale features of the reference subject are first extracted by a visual tokenizer. Instead of interleaving these conditional features with target image tokens across scales, our DreamVAR pre-fills the full subject feature sequence prior to predicting target image tokens. This design simplifies autoregressive dependencies and mitigates the train-test discrepancy in multi-scale conditioning scenario within the VAR paradigm. DreamVAR further incorporates reinforcement learning to jointly enhance semantic alignment and subject consistency. Extensive experiments demonstrate that DreamVAR achieves superior appearance preservation compared to leading diffusion-based methods.

cs.CV↗

arXiv2026

ToolACE-R: Model-aware Iterative Training and Adaptive Refinement for Tool Learning

Tool learning, which allows Large Language Models (LLMs) to leverage external tools for solving complex user tasks, has emerged as a promising avenue for extending model capabilities. However, existing approaches primarily focus on data synthesis for fine-tuning LLMs to invoke tools effectively, largely ignoring how to fully stimulate the potential of the model. In this paper, we propose ToolACE-R, a novel framework that includes both model-aware iterative training and adaptive refinement for tool learning. ToolACE-R features a model-aware iterative training procedure that progressively adjust training samples based on the model's evolving capabilities to maximize its potential. Additionally, it incorporates self-refinement training corpus which emphasizes LLM's ability to iteratively refine their tool calls, optimizing performance without requiring external feedback. Furthermore, we introduce adaptive self-refinement mechanism for efficient test-time scaling, where the trained model can autonomously determine when to stop the process based on iterative self-refinement. We conduct extensive experiments across several benchmark datasets, showing that ToolACE-R achieves competitive performance compared to advanced API-based models. The performance of tool invocation can be further improved efficiently through adaptive self-refinement. These results highlight the effectiveness and generalizability of ToolACE-R, offering a promising direction for more efficient and scalable tool learning.

cs.CL↗

arXiv2025

Training Report of TeleChat3-MoE

TeleChat3-MoE is the latest series of TeleChat large language models, featuring a Mixture-of-Experts (MoE) architecture with parameter counts ranging from 105 billion to over one trillion,trained end-to-end on Ascend NPU cluster. This technical report mainly presents the underlying training infrastructure that enables reliable and efficient scaling to frontier model sizes. We detail systematic methodologies for operator-level and end-to-end numerical accuracy verification, ensuring consistency across hardware platforms and distributed parallelism strategies. Furthermore, we introduce a suite of performance optimizations, including interleaved pipeline scheduling, attention-aware data scheduling for long-sequence training,hierarchical and overlapped communication for expert parallelism, and DVM-based operator fusion. A systematic parallelization framework, leveraging analytical estimation and integer linear programming, is also proposed to optimize multi-dimensional parallelism configurations. Additionally, we present methodological approaches to cluster-level optimizations, addressing host- and device-bound bottlenecks during large-scale training tasks. These infrastructure advancements yield significant throughput improvements and near-linear scaling on clusters comprising thousands of devices, providing a robust foundation for large-scale language model development on hardware ecosystems.

cs.CL↗

arXiv2025

Fine-grained Image Retrieval via Dual-Vision Adaptation

Fine-Grained Image Retrieval~(FGIR) faces challenges in learning discriminative visual representations to retrieve images with similar fine-grained features. Current leading FGIR solutions typically follow two regimes: enforce pairwise similarity constraints in the semantic embedding space, or incorporate a localization sub-network to fine-tune the entire model. However, such two regimes tend to overfit the training data while forgetting the knowledge gained from large-scale pre-training, thus reducing their generalization ability. In this paper, we propose a Dual-Vision Adaptation (DVA) approach for FGIR, which guides the frozen pre-trained model to perform FGIR through collaborative sample and feature adaptation. Specifically, we design Object-Perceptual Adaptation, which modifies input samples to help the pre-trained model perceive critical objects and elements within objects that are helpful for category prediction. Meanwhile, we propose In-Context Adaptation, which introduces a small set of parameters for feature adaptation without modifying the pre-trained parameters. This makes the FGIR task using these adjusted features closer to the task solved during the pre-training. Additionally, to balance retrieval efficiency and performance, we propose Discrimination Perception Transfer to transfer the discriminative knowledge in the object-perceptual adaptation to the image encoder using the knowledge distillation mechanism. Extensive experiments show that DVA has fewer learnable parameters and performs well on three in-distribution and three out-of-distribution fine-grained datasets.

cs.CV↗