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.

On the Practice of Scaling Search Conversion Rate Prediction

Scaling a Search Conversion Rate (CVR) prediction model, especially in high-traffic environments, presents a challenge: superior model quality needs to be balanced with strict constraints on training cost and serving latency. This paper details an effective approach for scaling modern search CVR prediction models. We begin with an empirical study to understand the scaling performance of search CVR models, analyzing how quality improves as we scale three key factors of model backbone computation, the size of embedding parameters, and the volume of training data. We use a large-scale production dataset, comprising over a year of customer interaction logs from a high-traffic e-commerce platform, to evaluate the scalability of several state-of-the-art architectures and their ensembles. Our key findings are: (1) selecting the right backbone and scaling factors is crucial; (2) the impact of scaling backbone, embedding, and data is largely independent and additive, which has implications for more efficient scaling exploration; (3) a streamlined warmstart strategy can accelerate training iterations while simplifying new updates; (4) inference optimization strategies such as decoupled graph execution and dynamic batching can enable low-latency GPU serving even for high-capacity models. Compared to a baseline of a pre-scaling production model, we ultimately deployed a model trained on 2.5x larger training data with 8x more inference compute while having minimal latency impact. Online A/B tests also demonstrate that our launches achieved a combined +2.6% gain in a key metric of search conversion rate.

PetroBench: A Benchmark for Large Language Models in Petroleum Engineering

Large Language Models are increasingly applied in the petroleum industry, highlighting the need for a domain-specific evaluation framework. This study develops a benchmark for LLMs in petroleum engineering, including a three-stage process of data preprocessing, quality filtering, and multi-model validation. Using expert review, a standardized question bank with strong domain relevance and discriminative capability was constructed. The benchmark covers production, reservoir, and drilling engineering, with 1,200 questions across multiple-choice, true or false, term definition, and short-answer formats. Eight mainstream LLMs were evaluated under a unified API environment. Results show that models performed better on subjective than objective questions, indicating weaknesses in factual knowledge discrimination. The highest accuracies for multiple-choice and true or false questions were 65.3% and 74.3%, respectively. Gemini-3-Pro, Kimi-K2.5, and Claude-Opus-4.6-Thinking achieved the best overall scores of 72%-74%. Models performed best in production engineering and weakest in reservoir engineering. Chinese models showed advantages in multiple-choice questions, while international models performed slightly better in short-answer questions. The benchmark provides a reproducible and practical reference for evaluating and deploying LLMs in petroleum engineering.

ReCoVR: Closing the Loop in Interactive Composed Video Retrieval

Composed video retrieval (CoVR) searches for target videos using a reference video and a modification text, but existing methods are restricted to a single interaction round and cannot support the progressive nature of real-world visual search. To bridge this gap, we first formalize interactive composed video retrieval, a multi-turn extension of CoVR, where users progressively refine their search intent through natural-language feedback across turns. Adapting existing interactive retrieval methods to this setting reveals two structural weaknesses: reliance on a single retrieval channel and an open-loop retrieval design that consumes user feedback but does not diagnose whether its own retrieval trajectory is drifting or stagnating. To address these limitations, we propose ReCoVR (Reflexive Composed Video Retrieval), a dual-pathway architecture built on reflexive perception, where the system treats its retrieval history as diagnostic evidence alongside user feedback. Specifically, an Intent Pathway routes heterogeneous feedback to complementary retrieval channels, while a Reflection Pathway performs trajectory-level reflection to monitor result evolution and correct retrieval errors across turns. Experiments on multiple benchmarks show that ReCoVR consistently outperforms interactive baselines, notably achieving 74.30% R@1 after just one interactive round on the WebVid-CoVR-Test dataset.

OpenGaFF: Open-Vocabulary Gaussian Feature Field with Codebook Attention

Understanding open-vocabulary 3D scenes with Gaussian-based representations remains challenging due to fragmented and spatially inconsistent semantic predictions across multi-view observations. In this paper, we present OpenGaFF, a novel framework for open-vocabulary 3D scene understanding built upon 3D Gaussian Splatting. At the core of our method is a Gaussian Feature Field that models semantics as a continuous function of Gaussian geometry and appearance. By explicitly conditioning semantic predictions on geometric structure, this formulation strengthens the coupling between geometry and semantics, leading to improved spatial coherence across similar structures in 3D space. To further enforce object-level semantic consistency, we introduce a structured codebook that serves as a set of shared semantic primitives. Furthermore, a codebook-guided attention mechanism is proposed to retrieve language features via similarity matching between query embeddings and learned codebook entries, enabling robust open-vocabulary reasoning while reducing intra-object feature variance. Extensive experiments on standard 2D and 3D open-vocabulary benchmarks demonstrate that our method consistently outperforms prior approaches, achieving improved segmentation quality, stronger 3D semantic consistency and a semantically interpretable codebook that provides insight into the learned representation.

Not all tokens contribute equally to diffusion learning

With the rapid development of conditional diffusion models, significant progress has been made in text-to-video generation. However, we observe that these models often neglect semantically important tokens during inference, leading to biased or incomplete generations under classifier-free guidance. We attribute this issue to two key factors: distributional bias caused by the long-tailed token frequency in training data, and spatial misalignment in cross-attention where semantically important tokens are overshadowed by less informative ones. To address these issues, we propose Distribution-Aware Rectification and Spatial Ensemble (DARE), a unified framework that improves semantic guidance in diffusion models from the perspectives of distributional debiasing and spatial consistency. First, we introduce Distribution-Rectified Classifier-Free Guidance (DR-CFG), which regularizes the training process by dynamically suppressing dominant tokens with low semantic density, encouraging the model to better capture underrepresented semantic cues and learn a more balanced conditional distribution. This design mitigates the risk of the model distribution overfitting to tokens with low semantic density. Second, we propose Spatial Representation Alignment (SRA), which adaptively reweights cross-attention maps according to token importance and enforces representation consistency, enabling semantically important tokens to exert stronger spatial guidance during generation. This mechanism effectively prevents low semantic-density tokens from dominating the attention allocation, thereby avoiding the dilution of the spatial and distributional guidance provided by high semantic-density tokens. Extensive experiments on multiple benchmark datasets demonstrate that DARE consistently improves generation fidelity and semantic alignment, achieving significant gains over existing approaches.

Efficient Camera Pose Augmentation for View Generalization in Robotic Policy Learning

Prevailing 2D-centric visuomotor policies exhibit a pronounced deficiency in novel view generalization, as their reliance on static observations hinders consistent action mapping across unseen views. In response, we introduce GenSplat, a feed-forward 3D Gaussian Splatting framework that facilitates view-generalized policy learning through novel view rendering. GenSplat employs a permutation-equivariant architecture to reconstruct high-fidelity 3D scenes from sparse, uncalibrated inputs in a single forward pass. To ensure structural integrity, we design a 3D-prior distillation strategy that regularizes the 3DGS optimization, preventing the geometric collapse typical of purely photometric supervision. By rendering diverse synthetic views from these stable 3D representations, we systematically augment the observational manifold during training. This augmentation forces the policy to ground its decisions in underlying 3D structures, thereby ensuring robust execution under severe spatial perturbations where baselines severely degrade.

XIT: Exploration and Exploitation Informed Trees for Active Gas Distribution Mapping in Unknown Environments

Mobile robotic gas distribution mapping (GDM) provides critical situational awareness during emergency responses to hazardous gas releases. However, most systems still rely on teleoperation, limiting scalability and response speed. Autonomous active GDM is challenging in unknown and cluttered environments, because the robot must simultaneously explore traversable space, map the environment, and infer the gas distribution belief from sparse chemical measurements. We address this by formulating active GDM as a next-best-trajectory informative path planning (IPP) problem and propose XIT (Exploration-Exploitation Informed Trees), a sampling-based planner that balances exploration and exploitation by generating concurrent trajectories toward exploration-rich goals while collecting informative gas measurements en route. XIT draws batches of samples from an Upper Confidence Bound (UCB) information field derived from the current gas posterior and expands trees using a cost that trades off travel effort against gas concentration and uncertainty. To enable plume-aware exploration, we introduce the gas frontier concept, defined as unobserved regions adjacent to high gas concentrations, and propose the Wavefront Gas Frontier Detection (WGFD) algorithm for their identification. High-fidelity simulations and real-world experiments demonstrate the benefits of XIT in terms of GDM quality and efficiency. Although developed for active GDM, XIT is readily applicable to other robotic information-gathering tasks in unknown environments that face the exploration and exploitation trade-off.

End-to-End Differentiable Photon Counting CT

Quantitative imaging is an important feature of spectral X-ray and CT systems, especially photon-counting CT (PCCT) imaging systems, which is achieved through material decomposition (MD) using spectral measurements. In this work, we present a novel framework that makes the PCCT imaging chain end-to-end differentiable (differentiable PCCT), with which we can leverage quantitative information in the image domain to enable cross-domain learning and optimization for upstream models. Specifically, the material decomposition from maximum-likelihood estimation (MLE) was made differentiable based on the Implicit Function Theorem and inserted as a layer into the imaging chain for end-to-end optimization. This framework allows for an automatic and adaptive solution of a wide range of imaging tasks, ultimately achieving quantitative imaging through computation rather than manual intervention. The end-to-end training mechanism effectively avoids the need for direct-domain training or supervision from intermediate references as models are trained using quantitative images. We demonstrate its applicability in two representative tasks: correcting detector energy bin drift and training an object scatter correction network using cross-domain reference from quantitative material images.

CUA-Skill: Develop Skills for Computer Using Agent

Computer-Using Agents (CUAs) aim to autonomously operate computer systems to complete real-world tasks. However, existing agentic systems remain difficult to scale and lag behind human performance. A key limitation is the absence of reusable and structured skill abstractions that capture how humans interact with graphical user interfaces and how to leverage these skills. We introduce CUA-Skill, a computer-using agentic skill base that encodes human computer-use knowledge as skills coupled with parameterized execution and composition graphs. CUA-Skill is a large-scale library of carefully engineered skills spanning common Windows applications, serving as a practical infrastructure and tool substrate for scalable, reliable agent development. Built upon this skill base, we construct CUA-Skill Agent, an end-to-end computer-using agent that supports dynamic skill retrieval, argument instantiation, and memory-aware failure recovery. Our results demonstrate that CUA-Skill substantially improves execution success rates and robustness on challenging end-to-end agent benchmarks, establishing a strong foundation for future computer-using agent development. On WindowsAgentArena, CUA-Skill Agent achieves state-of-the-art 57.5% (best of three) successful rate while being significantly more efficient than prior and concurrent approaches. The project page is available at https://microsoft.github.io/cua_skill/.