探索全球前沿学术脉络

UltraQuant: 4-bit KV Caching for Context-Heavy Agents

arXiv:2606.20474v1 Announce Type: cross Abstract: Context-heavy agents place unusual pressure on the key-value (KV) cache: long prefixes are reused across many short turns, while concurrency determines whether the serving system can keep GPUs utilized. We study 4-bit KV-cache compression for this setting, using TurboQuant-style rotation and codebook quantization as a quality anchor and vLLM FP8 KV caching as the deployment anchor. We report three contributions. First, we frame 4-bit KV caching around multi-round agent workloads where task quality, cache residency, and serving throughput must be measured jointly. Second, we describe the practical design choices needed to make the 4-bit path robust, including asymmetric K/V treatment, Walsh-Hadamard rotation, QJL removal, and block-scale variants. Third, we present serving optimizations on AMD GPUs, including optimized decode-attention kernels and UltraQuant, an FP4 approximation path that uses FP8 queries, FP4 KV tensors, UE8M0 group scales, and native scaled-MFMA support on CDNA4. On a long-context, multi-turn agentic workload, UltraQuant cuts P50 time-to-first-token by 3.47x in the cache-pressured late rounds (2.3x across all rounds) and raises output throughput by 1.63x over the FP8 KV baseline.

Variable-Width Transformers

Scaling model size, specifically depth and width, has driven significant progress in transformer-based language models. However, most architectures maintain a constant width across all layers, allocating a fixed parameter and computation budget evenly despite different layers potentially playing distinct computational roles. In this work, we empirically investigate nonuniform capacity allocation across network depth by proposing a $\times$-shaped >

Are Online Skill and Memory Modules Always Worth Their Tokens? A Budget-Constrained Study of Web Agents

Online web agents often augment a base actor with memory, workflow, or skill modules. These modules can improve performance, but they also consume test-time tokens, a cost rarely reported alongside the actor's inference cost. We study online augmentation, where this overhead is paid on every task, and re-evaluate its benefits under a fixed total inference budget. We compare AWM, ASI, and ReasoningBank with a token-matched vanilla baseline that uses the same budget for additional actor steps. Across three WebArena domains and three models, Gemini 3 Flash, GPT-5.4-mini, and Qwen 3.6-27B, the vanilla baseline matches or surpasses all three augmentation methods in aggregate success rate while often using fewer total tokens. We observe a similar trend on WorkArena-L1 with Qwen 3.6-27B, indicating that the effect extends to enterprise knowledge-work tasks. Our results suggest that skills and workflow memory can be useful in specific domains, but their apparent gains often vanish against a budget-matched actor. We further show that run-to-run variance materially affects outcomes and should be reported as a core evaluation criterion for online web agents.

Immunologically Optimized Zmp1 Peptides Reveal a Translational Serological Biomarker Platform for Tuberculosis Diagnosis Across Disease Manifestations

Tuberculosis (TB) diagnosis remains challenging, particularly for extrapulmonary TB (EPTB), where invasive sampling, low bacillary burden, and suboptimal sensitivity of nucleic acid-based tests in peripheral specimens hinder timely detection. Here, we report an immunology-driven strategy for biomarker discovery and development of a peptide-based serological assay targeting Mycobacterium tuberculosis zinc metalloprotease-1 (Zmp1). Leveraging fundamental principles of adaptive immunity that antigenic regions containing overlapping B-cell and CD4 T-helper cell epitopes would preferentially generate high antibody titers through linked recognition and cognate T-cell help, we used an immunoinformatics pipeline to identify two nested immunodominant peptide regions within Zmp1 (Mtb-Zp-NT and Mtb-Zp-CT) enriched for overlapping B- and T-cell epitopes. The diagnostic potential of these peptides was evaluated through ELISA-based serological assays. A blinded pilot study (N=137) demonstrated a clear discrimination between active TB and TB-recovered individuals. The assay was subsequently validated in an expanded cohort (N=875) by screening 6,086 individuals, which identified 457 TB-positive cases. The cohort included pulmonary TB (PTB), EPTB, TB-recovered individuals, household contacts, non-specific infections, and healthy controls. Receiver operating characteristic analyses, supported by DeLong and bootstrap comparisons, revealed superior diagnostic performance of the peptide-based assays relative to full-length Zmp1. Mtb-Zp-CT exhibited the highest accuracy (AUC=0.93; specificity >90%), while Mtb-Zp-NT also demonstrated strong discriminatory power (AUC{approx}0.89). These findings establish that the immunologically optimized Zmp1 peptides are highly promising serological biomarkers for TB and EPTB. More broadly, they demonstrate how mechanistically informed epitope selection can accelerate translation of pathogen-specific immune signatures into sensitive, minimally invasive, and potentially point-of-care diagnostic platforms for resource-limited settings.

Grounding Computer Use Agents on Human Demonstrations

arXiv:2511.07332v2 Announce Type: replace-cross Abstract: Building reliable computer-use agents requires grounding: accurately connecting natural language instructions to the correct on-screen elements. While large datasets exist for web and mobile interactions, high-quality resources for desktop environments are limited. To address this gap, we introduce GroundCUA, a large-scale desktop grounding dataset built from expert human demonstrations. It covers 87 applications across 12 categories and includes 56K screenshots, with every on-screen element carefully annotated for a total of over 3.56M human-verified annotations. From these demonstrations, we generate diverse instructions that capture a wide range of real-world tasks, providing high-quality data for model training. Using GroundCUA, we develop the GroundNext family of models that map instructions to their target UI elements. At both 3B and 7B scales, GroundNext achieves state-of-the-art results across five benchmarks using supervised fine-tuning, while requiring less than one-tenth the training data of prior work. Reinforcement learning post-training further improves performance, and when evaluated in an agentic setting on the OSWorld benchmark using o3 as planner, GroundNext attains comparable or superior results to models trained with substantially more data,. These results demonstrate the critical role of high-quality, expert-driven datasets in advancing general-purpose computer-use agents.