每天知道
你的领域
出了什么

AI agents can automate tasks by turning a single natural-language request into a multi-step process spanning tools, files, and applications. Users are often left to judge that process from fragmented execution information and the final output. To make the completed process easier to understand, validate, and reuse, we investigate post-task workflows: editable, graph-based representations of an agent's completed execution. We first analyzed 10,803 public workflow templates from n8n to characterize real-world automation practice, then developed Trace2Flow, a research probe that translates agent execution traces into interactive post-task workflows. In a study, participants (N = 20) reviewed agent executions with prompt or agent errors. We found that post-task workflows improved their understanding and error detection over a prompt-only condition, and that validation succeeded mainly when users cross-checked across multiple evidence sources. For follow-up tasks, adapting the workflow matched adapting the prior prompt in success, time, and difficulty, and was often preferred.

We demonstrate that a general-purpose agent can directly drive a physical robot throughout task execution without any task-specific or environment-specific training. We introduce Agent as Policy (AGP), which places task planning and execution under the agent's control. Given a task and a robot interface, the agent interprets visual evidence, writes executable programs, issues motion commands, and revises its actions in response to physical outcomes. This brings the agent's reasoning and programming capabilities into continuous interaction with the physical world. We study AGP across multiple real-world manipulation tasks spanning precision manipulation, dynamic motions, and deformable objects. These include assembly from human videos, block construction from goal images, dice flipping, targeted throwing, and bimanual towel folding. Across assembly, block construction, and dice flipping, AGP succeeds in at least eight of ten trials for each evaluated task configuration. We further study efficiency through task experience accumulation and find that reusing saved procedures and programs shortens execution time across repeated trials. These findings support a path for general-purpose agents to act as robot policies, extending their autonomy to physical manipulation through runtime reasoning, programming, and interaction.

Mobile GUI agents are shifting from multi-module frameworks to native models trained end-to-end, yet industrial deployment faces three persistent gaps. Sandbox training produces a distribution mismatch with production environments; expensive real-device failures remain underutilized; and fixed benchmarks saturate, losing the power to guide iteration. We present BlueLM-GUI, a 35B-A3B mobile GUI agent built as a real-device-centric flywheel that closes these gaps through three principles. Every Sample Matters: a dual-track pipeline with Heterogeneous Triple-System Consensus evaluation and an Error Correction & Derivation Module salvages every trajectory into usable supervision. Every Rollout Is Real: a three-stage recipe---continual pre-training, supervised fine-tuning, and agentic reinforcement learning on hundreds of real phones---grounds every rollout in real production environments, so the capability the model learns transfers directly to deployment. Every Query Evolves: a quota-driven benchmark methodology with three orthogonal axes enables precise attribution and allows the benchmark to be systematically upgraded as the model improves. BlueLM-GUI achieves 87.4 on MobileGUI-VBench, surpassing the best closed-source model by 5.1 points, and 84.9 on AndroidWorld, the best result among open-source models and competitive with closed-source models. These results demonstrate that grounding model training and iterative improvement in both real devices and the three Every principles yields strong, robust, and transferable mobile GUI capability.

Web agents can utilize reusable tools to reduce the cost and latency of low-level browser interaction, but automatically discovered tool collections can be large, redundant, and poorly aligned with user demand. We present AutoTailor, a meta-agentic framework for constructing and maintaining a compact set of trajectory-derived Model Context Protocol (MCP) APIs. Offline, AutoTailor converts web trajectories into parameterized browser-automation programs, applies a Quality Filter to remove APIs with unsuitable granularity and redundant functionality, and applies a Usage Likelihood Filter to prioritize broadly useful capabilities while preserving semantic coverage. Online, Dynamic Reselection monitors task outcomes and API usage, identifies recurring coverage gaps, adds relevant candidates, and prunes persistently unused capabilities. We evaluate AutoTailor on 106 WebArena Postmill tasks. Offline filtering reduces the initial 1,283 unrefined APIs to 87, and Dynamic Reselection produces a 33-API set. With reasoning and acting (ReAct) fallback, this set achieves 90.6% correctness, compared with 87.5% for ReAct alone, while reducing average total request-token cost by 57.8% and latency by 29.4%. Without ReAct, it achieves 60.1% correctness, marginally matching the performance of unrefined set, while reducing request-token usage by 94.9%. Together, these results show that static filtering produces a compact inventory of APIs expected to support core, high-likelihood tasks, while dynamic reselection further tailors that inventory to observed user needs. This combination improves accuracy and latency while sharply reducing token usage and end-to-end cost, demonstrating the value of user-aligned capability management for efficient web agents.

Enterprise AI agents often succeed in a demonstration and then stall once they must operate day after day. An industry report estimates that most pilots never reach production and that deployed systems rarely retain feedback or improve over time, while agent benchmarks show single-run successes masking unreliable repetition. We argue that these failure modes share a common architectural root: agents reason over data structured for human operators and traditional applications, not for the language models that power them. This position paper proposes that companies deploying agents in sustained operation should rebuild their cognitive substrate, the shared environment agents read as working context, around representations matched to that reasoning surface, isolating schema translation to the action boundary. Markdown is the instantiation available today, not a proven agent-native primitive. Two mechanisms ground the argument: context-bandwidth asymmetry, the gap between one-pass reading of connected prose and field-by-field typed access that strips relations; and cross-loop coupling, the claim that action, skill, and policy loops compound only if they share one substrate. A four-layer framework (Data, Knowledge, Intelligence, Governance) operationalizes the position, with a Sync Agent enforcing the action boundary and a per-skill trust gradient, making governance and auditability structural properties of the substrate. The position revives the shared-substrate tradition of classical multi-agent systems under LLM-era economics. We analyze the main objections and risks, including indirect prompt injection on the compile path, and outline a research agenda for evaluating substrates directly.

Large language models (LLMs) are rapidly evolving from conversational assistants into agentic AI systems that reason, plan, invoke tools, maintain persistent memory, communicate with other agents, and execute multi-step tasks. At the same time, modern models exhibit substantially stronger native safety alignment than earlier generations on which many jailbreak attacks and defenses were originally studied. This shift raises a fundamental question: which established jailbreak-security findings remain valid in the era of modern LLMs and agentic AI? We address this question through a Systematization of Knowledge (SoK) that reframes jailbreak security around the full agentic execution pipeline. We develop unified taxonomies of attacks and defenses spanning user interaction, planning and reasoning, memory, tool use, and inter-agent communication, and introduce a security--utility--efficiency evaluation framework that separates native harmful-prompt safety, adversarial jailbreak robustness, and agent-level security outcomes. We further conduct a controlled empirical study of representative attacks and defenses within a common agentic framework. Our results reveal three important gaps. First, strong native alignment does not imply robustness to adversarial jailbreaks. Second, defense effectiveness is highly model-, attack-, and component-dependent and can come at substantial cost in over-refusal, utility, and latency. Third, low final-response attack success can mask severe intermediate compromise: planning, memory, and tool interactions may remain unsafe even when the final response is successfully filtered. These findings motivate a shift from response-centric jailbreak defense toward cross-layer, execution-aware security that protects agent state, component transitions, and external actions while preserving practical utility and efficiency.

Large language models are being organized into multi-agent systems with specialized roles, but whether such specialization produces distinct forecasts and whether subsequent synthesis improves utility remains unclear. In this study, we carried out a live, prospective evaluation over the final 56 matches of the information-dense 2026 FIFA World Cup, keeping a frontier foundation model constant while assigning two primary forecasting agents contrasting specialist roles: a quantitative specialist focusing on structured performance statistics and a news specialist focusing on current injuries, tactics and information from press conferences. Their forecasts were then reviewed by a separate critic before being combined by a meta-agent, resulting in a sequential four-agent model. Forecasts from the betting market served as an external benchmark. The news specialist obtained the highest mean probability-weighted Top-3 utility and matched the betting market in Top-3 exact-score hits. Nevertheless, the two specialist forecasters agreed on at least two of the three scorelines in 50 out of 56 matches, and the meta-agent never generated more than one scoreline outside the specialists' forecast set. These findings show that rapidly changing, unstructured information can provide a valuable forecasting signal alongside structured statistics, whereas adding critic and meta-agent stages does not necessarily create complementary information or improve on the strongest specialist.

Existing agent benchmarks mainly evaluate final task success or tool-call correctness, providing limited insight into whether agents can reliably diagnose and recover from intermediate execution failures. This limitation becomes particularly critical in multi-turn parallel tool-use scenarios, where errors may propagate across dependent branches and trigger cascading failures. We introduce ParaRecover, a process-level benchmark for evaluating error localization and recovery in multi-turn parallel tool-use agents. Built upon a fine-grained taxonomy of 14 error types covering planning dependencies, tool selection, and argument matching, the benchmark comprises 10,626 instances spanning two difficulty levels. To enable finegrained, process-oriented evaluation, we further propose the SDE rubric, which measures structural integrity, diagnostic reasoning, and evolutionary strategy during agent execution.Experiments across more than ten mainstream LLMs reveal that even state-of-the-art models still struggle with multi-turn error propagation,implicit tool-use failures, and precise replanning. Moreover, we demonstrate that the SDE rubric provides effective supervision signals for improving agents' reflective recovery capabilities. Our data and code are available at https://github.com/gbw206/ParaRecover.

Lux AI Season 3 requires agents to act under partial observability, randomized episode level dynamics, and a best of five match structure that rewards both tactical execution and fast adaptation. We present HORIZON, a hierarchical agent that combines symmetry aware spatial perception, dual memory belief tracking, relic centric graph attention, information gain driven exploration, and an opponent conditioned policy mixture. HORIZON separates short horizon control from cross match meta reasoning, while auxiliary belief and world model objectives stabilize learning. Trained with PPO in a large scale JAX simulator, the resulting agent explicitly infers hidden game parameters and opponent style. Experiments show consistent gains in match win rate, episode win rate, adaptation gain, and league rating over strong recurrent and feed forward baselines.

Large Language Models (LLMs) have enabled more natural human-robot interaction, but open-source models often exhibit unstable long-horizon reasoning and inefficient action execution when deployed in agentic robotic frameworks. This paper presents an enhanced ROS-Agent based architecture that improves task reliability and execution efficiency for agentic robotic systems using open-source LLMs. The proposed system introduces a novel intermediate mechanism, termed the MetaTool, which enforces structured planning prior to action execution. Given a natural-language command, the MetaTool induces the LLM to generate a pseudo-code plan of intended tool invocations, which is stored in the ROS-Agent's scratchpad and persists throughout execution. By explicitly separating planning from execution, the proposed approach reduces execution loops and improves deterministic behavior. The architecture is validated on a custom mobile robotic platform with multimodal perception and motion control capabilities. Experimental results on real-world interactive tasks demonstrate improved task completion and contextual consistency, with up to ~24% gains on complex tasks compared to the baseline framework.

Real-world Earth observation (EO) agents must translate high-level scientific questions into executable workflows to acquire observations, prepare data, perform domain computations, and derive conclusions from runtime evidence. Existing EO agents typically start from supplied observations, while benchmarks typically provide prepared inputs or candidate answers, leaving full-chain open-world EO execution largely untested. We present Earth-Agent-Pro, an execution-adaptive Plan-and-Execute framework using expert-authored skills to constrain planning and runtime tool use. Workflow-centered structured memory records planned steps, accepted evidence, and their dependencies, enabling repair of only the affected workflow suffix when runtime evidence invalidates a step. Separate large language model adapters use sequence-level supervised fine-tuning for planner workflow composition and node-level group relative policy optimization with locally verifiable rewards for executor tool-argument grounding. Earth-Bench-Pro instantiates 248 expert-curated task cores as 744 questions under three matched regimes. Its 248 Open-World Execution questions span RGB imagery, spectral observations, and remote sensing products, pairing high-level requests with runtime data requirements, executable trajectories, and open-ended answers grounded in execution evidence. With a shared GPT-5 backbone, Earth-Agent-Pro achieves 66.13% LLM-as-Judge accuracy, exceeding ReAct by 20.95 points in this metric and 24.44 points in Tools-In-Order. Joint adapter tuning raises Qwen3.5-9B LLM-as-Judge accuracy from 38.31% to 50.00%, an 11.69-point gain over the untuned configuration. Planning-only evaluation and execution with the reference workflow show that the adapters improve workflow composition and argument grounding, respectively. Code and datasets will be released soon.

Large Language Model (LLM) agents offer a promising path toward autonomously managing long-term physical tasks without human intervention. However, physical tasks require agents to continuously observe the environment, make consequential actions, and remain effective as the environment changes. Existing approaches either require substantial data and retraining, or primarily focus on agents operating in the virtual world. In this work, we explore the feasibility of building a self-adaptive physical AI agent that manages long-term physical tasks in a zero-shot manner and adapts to environmental changes without human intervention. We design a multi-agent framework that integrates planning, tool calling, observation, and verification, and evaluate it on agricultural tasks against reinforcement learning (RL) agents under different weather patterns. Our results show that zero-shot LLM agents can achieve comparable management outcomes to RL agents under the same weather pattern and adapt more effectively than RL when evaluated under a shifted environment, highlighting a promising path toward self-adaptive physical AI agents.

Traditional large language models (LLMs) are scoped to individual user sessions, limiting their knowledge to a single conversation and preventing them from learning user preferences that evolve over time. Existing agentic memory systems address this limitation but generally operate at the individual-user level, restricting the public knowledge that could be shared across users to improve downstream responses. We introduce AIM (Agentic Interoperable Memory), a unified, privacy-aware memory framework that enables multi-agent, multi-user LLM systems to persistently manage private and shared memory. AIM dynamically classifies information as private, scoped to one user and inaccessible to others, or public, accessible to all users. It enforces index-level access controls so that private memories are retrievable only by their owner, protecting sensitive data while allowing beneficial shared knowledge to improve coordination and consistency. We also introduce MUMBench (Multi-User Memory Benchmark), a dataset of multi-user interactions containing private and shareable information across four domains. To our knowledge, MUMBench is the first public dataset designed to evaluate multiple memory operations, including retrieval, creation, update, and deletion, in a multi-user environment. Across three independent runs on MUMBench, AIM achieves 96.0% visibility classification accuracy, 58.8% strict operation accuracy, and 70.5% state-aware operation accuracy.

Real-time motion planning must execute under strict latency and energy constraints on resource-limited platforms. This paper presents Arborist, an algorithm-hardware co-design framework that accelerates Fast Marching Tree (FMT*) motion planning through synergistic algorithmic and architectural innovations. At the algorithm level, Arborist introduces safe multi-tree expansion with global competition management and cross-tree optimality checking to expose inter-tree parallelism while preserving path quality. At the architecture level, our proposed ArboristAccel integrates an Arborist Toolbox that provides hardware support for inter-tree parallelism; a KD-tree Subsystem for high-throughput near-neighbor search; a Parallel-Query Retrieval module that balances multi-bank memory accesses to alleviate system-level memory bandwidth bottlenecks; and a Look-ahead Planning Engine that exploits intra-tree parallelism while preserving path quality by in-order commit. Synthesized in 16 nm and operating at 500 MHz, ArboristAccel sustains more than 1000x speedup over CPU across all workloads, peaking at 2800x on 2D-Maze, and delivers geomean improvements of 8.8x speedup, 4.6x area efficiency, and 3.1x power efficiency over an FMT* ASIC baseline. The corresponding factors relative to a state-of-the-art RRT*-based MOPED accelerator are 8.6x, 1.0x, and 1.8x, at comparable path cost. To the best of our knowledge, ArboristAccel is the first dedicated hardware accelerator for FMT*-based motion planning.

Foundation models, including large language models, vision-language models, and time-series foundation models, are increasingly deployed on embedded and edge platforms for CPS and IoT applications, where energy, latency, and memory are as critical as task accuracy. Existing benchmarking tools evaluate model capability in isolation, reporting accuracy assuming sufficient compute, while hardware profiling tools remain platform-specific and mutually incompatible. As a result, users lack a unified workflow for making deployment decisions across heterogeneous devices. We present HoliBench, a modular benchmarking and deployment toolkit that jointly characterizes accuracy, latency, and energy across platforms from single-board computers to GPU servers. Its platform abstraction layer calibrates cross-device measurement, and the toolkit supports multiple model modalities, inference engines, concurrencies, and existing evaluation harnesses. An interactive interface exposes constraint-aware configuration selection over a design space that is profiled once and reused across studies. Using HoliBench, we characterize 20 models across 7 device types, 3 quantization levels, 8 inference backends, and over 30 tasks, surfacing tradeoffs that existing tools miss: quantization reduces latency only on hardware with low-precision support, accuracy gains show diminishing returns relative to energy, and for autoregressive workloads, average inference power is approximately constant across output lengths. We further find that single-model profiles compose under sequential co-resident execution. In a multi-model CPS deployment, standalone profiles predict combined-pipeline latency and power within 1.2% and 2.5%, enabling deployment exploration without exhaustively profiling every pipeline configuration. We release HoliBench as open-source infrastructure for deployment-aware evaluation of foundation models.

We introduce AMDKernelVault, an open HIP and Triton kernel corpus and training framework for recent AMD CDNA GPUs. Existing LLM-based kernel agents are largely CUDA/NVIDIA-centric and often depend on repeated frontier-LLM calls for generation, reflection, and optimization. To address this gap, we develop HIPKernelGen and TritonKernelGen, agent-driven pipelines that transform PyTorch references into HIP or Triton kernels, compile and validate candidates under ROCm, and latency-profile them on AMD hardware. The corpus contains 62,153 execution-verified HIP kernel samples, 2,377 production-grounded ROCm Libraries QA entries, and 39,893 Triton kernels. We further train Qwen3-8B with supervised fine-tuning and execution-aware reinforcement learning as a demonstration of the corpus's utility. Under fixed evaluation budgets, it achieves the highest correctness among the compared models on PyTorch-to-HIP (34.0% Pass@1), TritonBench-G (33.2% Corr@3), and ROCmBench (41.94% Corr@3), but does not uniformly lead compilation or speed metrics. The corpus and documentation are available at https://huggingface.co/datasets/amd/AIG-Datasets, and the associated training and kernel-generation code is available at https://github.com/AMD-AGI/hip_kernel_llm_lab.

Dense retrieval systems increasingly rely on multi-billion-parameter language models, whose memory and computational requirements make large-scale indexing, frequent corpus updates, and low-latency serving costly. We present a three-stage training pipeline for developing compact and efficient retrievers that remain competitive with substantially larger models. The pipeline combines cross-lingual alignment, relational knowledge distillation, and contrastive fine-tuning. It requires no original ground-truth relevance labels, relying exclusively on supervision generated by strong embedding models and rerankers utilised as teachers. Using this pipeline, we develop PolDense and EuroDense, both supporting contexts of up to 8,192 tokens. PolDense is a family of six Polish retrievers ranging from 17M to 1B parameters. EuroDense is a 435M-parameter retriever supporting nine European languages. We conduct an extensive evaluation covering 41 Polish and 150 multilingual retrieval tasks. The results demonstrate strong quality-efficiency trade-offs. PolDense-1B outperforms the evaluated retrievers with up to 9B parameters, while the PolDense family forms the Pareto frontier across model sizes. Among the evaluated models below 1B parameters, EuroDense ranks first in both task-averaged and language-averaged performance and leads in seven of nine languages. We release all models publicly.

How much learned memory is needed to benefit from more data? We show that the two resources are governed by one predictive-energy spectrum in a positive-entropy autoregressive retrieval source. Each coordinate contributes its query probability times the squared radius of its unknown logit. Writing $μ$ for the resulting energy spectrum, we prove the minimax law $\mathfrak R^*_{\rm value}(n,B)\asymp_R Φ_μ(n^{-1})+Φ_μ(τ_B), Φ_μ(t)=\int\min\{x,t\}\,μ(\mathrm dx),$ for $n$ prediction blocks and a learned state with at most $2^B$ values. Data set the resolution $1/n$; memory sets the level $τ_B$ reached by optimal bit allocation. The complete curve also recovers the positive spectrum. Energy-dimension pairing is essential: two causal sources with identical block-energy and block-dimension marginals have different data and memory exponents. A masked query-key attention head learns the route and values, realizing the law with explicit routing, format, and arithmetic errors. Further results give exponent-adaptive allocation, finite-precision realization, and compute-precision laws under two-sided arithmetic assumptions. Experiments recover the data-memory collapse and coupling exponents, explain the routing and allocation mechanisms, and examine weight-only quantization across six pretrained-model scales.

Efficient GPU implementations of tensor programs often require joint optimization of high-level algebraic formulations and low-level execution strategies. However, the resulting search space grows rapidly as transformations combine across operators, making joint optimization difficult to scale. We present EqiForge, a tensor program superoptimizer based on equality saturation. Its unified IR represents high-level tensor expressions and tiled computations in a single expression language. By composing equality rules, EqiForge derives fused implementations such as FlashAttention-style kernels directly from tensor expressions. Early compaction prunes redundant partial programs before completion, while subgraph composition extends the search to larger graphs. Across tensor-program benchmarks, EqiForge achieves a geometric mean speedup of 1.32x and a maximum of 2.74x over the fastest available baseline per configuration. Its attention kernels outperform FlashAttention by up to 1.87x in decode and approach its performance in prefill. EqiForge also discovers new implementations that outperform torch.compile on various Transformer layers, including QK-normalized MLA (3.16x) and mHC (5.84x).

Modern AI systems are built on the Transformer architecture, whose core operation, attention, accounts for the majority of computation and memory cost. Researchers continually propose new attention variants to improve quality, efficiency, or context length, but each variant currently requires expert-written GPU code to run at usable speeds. PyTorch's recent flex_attention lets researchers describe custom attention patterns in Python and compile them to fused kernels, but its design is limited to modifications applied after the central matrix multiplication, excluding Rotary Position Embedding (RoPE), the positional encoding used by every major LLM. We introduce AttnFuse, a small DSL for attention that makes pre-multiplication transformations like RoPE first-class operations. Researchers compose ten high-level building blocks to describe a variant, and AttnFuse's compiler emits a single fused GPU kernel for the entire computation. On an RTX 3090, AttnFuse achieves a 2.10$\times$ speedup over flex_attention on the RoPE+causal pattern. On an H100, it runs a full Llama-3-8B training step within 5% of PyTorch's hand-tuned backend. Our investigation reveals the Rotation Calculus: whether to fuse RoPE or apply it separately depends on the GPU's compute-to-bandwidth ratio, with a derived crossover that matches measurement. AttnFuse demonstrates that a small, attention-specific compiler can close the gap between flexible research code and production kernels.

Quantized Mixture-of-Experts (MoE) services can hold several pre-materialized instances of one base model, but quantization damage varies sharply across requests and bitwidths. Because instance materialization and replica counts consume memory and require slow reconfiguration, we treat them as upstream provisioning decisions and study routing within a fixed resident pool. Within this fixed-pool boundary, we route each request to maximize modeled throughput under a class-level expected quality-degradation budget and measured instance capacities. To predict this request-specific risk, we introduce FWP (Fragility-Weighted Perplexity), computed from prompt tokens on a reference-instance prefill and calibrated to candidate-instance degradation. Underlying FWP is an exact two-expert affinity--fragility decomposition and a conditional multi-layer top-$k$ expansion whose bias, interaction, route-change, separability, and higher-order terms remain explicit. Using these calibrated risks, a window-level linear program yields a signed reduced-reward score that is KKT-consistent with the LP optimum under optimal prices and primal-feasible tie allocation. On 88 extended Qwen prompts, complete W2, W3, and W4 instances quantizing all 6,144 expert blocks incur mean $Δ$NLL of $0.9437$, $0.1832$, and $0.0513$. Under the same population and $τ=0.1513$, FWP allocation reaches a $1.284\times$ offline model-based multiplier versus $1.253\times$ for request-agnostic mixing and $1.000\times$ for static W4, an incremental $2.5%$ relative FWP gain.

Post-training attention sparsification reduces the quadratic cumulative attention cost of pretrained Transformers by selecting a small set of context units (tokens or blocks) for each query. Existing trainable methods usually use a lightweight selector to score context units, followed by hard Top-K selection that blocks gradients from the language modeling loss. Consequently, these methods commonly distill layer-wise dense attention distributions. Although this encourages the selector to rank context units by dense attention weights in the original model, the ranking is not directly aligned with their impact on predictions under a fixed attention budget (i.e., the number of attended context units per query), potentially wasting the limited budget on less useful units. To address this misalignment, we propose Simple Attention Sparsification (SAS), a gated sparse attention mechanism that optimizes context ranking end-to-end with the language modeling loss. The key idea is to inject the selector's continuous scores into attention logits during training, allowing the loss to update the selector through standard backpropagation. We identify several choices crucial for this simple design to work well in practice: placing the gate inside the attention softmax in log form, using normalized softmax gates to calibrate historical context against the always-retained current block, and preserving continuous selector scores so the model learns relative priorities rather than only hard selections. To support long-sequence training, we implement a memory-efficient Triton kernel that integrates SAS into FlashAttention-style computation. Across reasoning, long-context understanding, and agentic tasks, SAS consistently outperforms trainable sparse attention baselines across attention budgets, with especially large gains under tight budgets, demonstrating more effective context ranking for downstream tasks.

A single SM utilization percentage can make an LLM inference workload look compute-saturated while hiding how much useful work is being done. The problem is not that the counter is wrong, but that it collapses several different mechanisms into one number. This is most severe during decode, where each request contributes only one new token and dense projection GEMMs become small-row matrix multiplications. On Hopper, the bfloat16 GMMA path executes these operations in fixed 64-row matrix fragments, so small-batch decode can fill only a small fraction of each fragment with real token rows. In this paper, we profile vLLM with FlashAttention-3 and cuBLASLt on an H100 NVL across cold prefill, warm prefill, and decode, sweeping sequence length and batch size. We replace the usual single utilization number with eight counter-validated views derived from raw Nsight Compute reports, each pinned to an NCU counter or explicit formula. Together, these views map utilization gaps to concrete mechanisms - fragment fill, occupancy limits, stall signatures, wave quantization, and kernel selection - across four production models and six per-layer kernel roles.

Frontier language models are more aggressively using subquadratic attention to reduce the memory footprint and compute requirements during inference while still delivering frontier accuracy. While existing systems make dense attention-centric disaggregated serving decisions, we show that disaggregating inference around the unique arithmetic intensity and memory footprint of subquadratic attention LLMs can achieve significant throughput and energy efficiency gains on emerging DRAM-based and SRAM-only heterogeneous systems. We introduce SQD (SubQuadratic Disaggregation), a fine-grained heterogeneous disaggregation scheme that splits decode by quadratic and subquadratic attention rather than by operator type, and that applies across subquadratic attention variants. For sparse attention LLMs, we disaggregate decode into top-k selection, which must index through the full KV, and top-k attention plus FFN, which have static memory footprints. For linear and sliding-window attention LLMs, we disaggregate decode into dense attention layers and subquadratic attention layers plus FFN. In an adjusted 8xB200 heterogeneous system proxy, we observe average tokens/J improvements of 53% on GLM 5.2, 31% on Nemotron 3 Ultra, and 56% on Gemma 4 31B over the strongest GPU-only baselines. In an analytical model of a Rubin plus LPX system with fixed power budgets, we observe 1.2x to 1.5x tighter achievable latencies and up to 3.6x higher throughput over the best baseline of attention-FFN disaggregation. Our experiments also reveal architectural insights on chip and interconnect provisioning for next-generation heterogeneous systems serving subquadratic attention.

GPU memory bandwidth and capacity limit throughput in large language model (LLM) inference. The GPU memory system consists of a primary tier of high-bandwidth memory (HBM) and a secondary tier of host memory connected via CPU-to-GPU interconnect. Current serving systems treat the tiers hierarchically: they serve exclusively from HBM when data fits, and otherwise prefetch data from host memory to HBM before use. In both cases, the host memory bandwidth is never well utilized. Prefetching expands capacity by utilizing host memory, but consumes HBM bandwidth for writes, reducing the bandwidth available for demand loads. We observe that fully utilizing both host and HBM bandwidth requires each wave of GPU threadblocks to access both tiers concurrently and in proportion to their bandwidth ratio. Existing bandwidth-proportional placement strategies fail to provide concurrency because they are not aware of GPU waves, and the large 2MB GPU page size. This paper presents BOOST, the first runtime system that provides concurrent and proportional access to both GPU memory tiers, extracting the combined bandwidth of host memory and HBM for LLM inference without kernel changes. The key insight in BOOST is to use kernel access patterns to make page allocation and runtime data management wave-aware. For static model weights, BOOST applies modulo-based page placement that eliminates access-ratio variance; for dynamically provisioned attention key-value (KV) pairs, it makes the free KV page pool wave-aware. We integrate BOOST into vLLM and evaluate on a Grace Hopper system. At iso-batch size, BOOST improves Time-per-Output-Token (TPOT) by 4.3% over HBM-only serving, whereas prefetching degrades TPOT by 6%. In high-throughput serving, BOOST improves throughput by 31% on average, outperforming prefetching by 15%.

AI is beginning to make substantive contributions to LLM inference optimization. Existing AI optimizations are predominantly profiling-based. Profiling-bound feedback confines the search to the capabilities and performance of an existing software stack, preventing a fundamentally better architecture of LLM inference systems from being identified. To enable the AI-driven LLM inference system architecting loop, we argue that a general workload representation, a verifiable mutation space, and an implementation-independent evaluator are required. We present the RoofLang domain-specific language (DSL) that provides these features. In our evaluation, RoofLang reveals that DeepSeek V4-series models could achieve 3.5-39.5$\times$ higher peak decode throughput than other representative models. This gap is disproportionate to their total parameter counts and arises largely from compact KV-cache designs that support larger batches and reduce memory traffic. A persistent optimizer agent further discovered several new architectures that improved both throughput and interactivity of DeepSeek V4 Pro on NVIDIA B300 by 6.23-50.1%.

With the recent rise and adoption of tabular foundation models, optimizing their inference performance becomes an emerging field for efficiency research. While the models are architecturally similar to transformer-based large language models (LLMs), the size and serving patterns differ significantly. We show that the focus should be on the attention calculation and less on weight or KV cache quantization, which are more popular in LLMs. We develop a quantization strategy for queries, keys, and values to FP8 and use explicit FP8 matrix multiplication instructions to speed up the attention calculation. We find that it is crucial to align the quantization error in the test rows with the quantization error in the training rows, as otherwise the accuracy drops drastically. Our Triton kernel achieves a speedup up to 1.7x over regular 16-bit kernels, and we show that on TabPFN-v3 and TabICLv2 there is no relevant accuracy loss across TabArena and BeyondArena.

Long-running multi-turn requests accumulate reusable key-value (KV) state. Once this state exceeds a fixed GPU KV-cache allocation, serving systems evict reusable prefixes, repeat prefill work, and may preempt requests. This pressure is particularly acute for Mixture-of-Experts (MoE) models: their expert weights occupy most GPU high-bandwidth memory (HBM), even though each token activates only a sparse subset of experts. A static boundary between weights and the KV cache prevents serving systems from using expert memory to preserve reusable state as conversations grow. We present VAMP, an MoE serving framework that changes this HBM boundary at runtime. When a KV allocation cannot be satisfied, VAMP compares the estimated future work of three alternatives: staging expert weights from host memory, evicting cached prefixes that may require re-prefill, or preempting and rescheduling requests. It then converts a bounded expert-weight region into KV-cache capacity when that action has the lowest estimated penalty. CUDA Virtual Memory Management page remapping performs this conversion without copying resident KV data. We implement VAMP in the vLLM serving engine and evaluate Qwen3-Next-80B on recorded and controlled multi-turn workloads. Across five replays of a recorded 2,103-turn SWE-bench agent workload, VAMP with a 15% maximum expert-offloading ratio reduces time-to-first-token (TTFT) p90 from 26.1 s to 1.10 s (23.6 times) and increases request throughput by 20.7% relative to unmodified vLLM, while increasing time per output token (TPOT) by 31.1%.

Test case synthesis is crucial for evaluating and ranking programs generated by large language models (LLMs). However, constructing high-quality test cases remains challenging because reliable expected outputs are often difficult to obtain. We propose Confidence-Gated Transductive Test Generation (CoTT), which first uses an efficient inductive procedure and invokes transductive generation only when inductive confidence is low. This adaptive design improves output reliability while allocating extra computation only when needed. On code reranking benchmarks, CoTT outperforms prior baselines across the reported metrics while reducing cost relative to applying transductive generation to every input. These results show that confidence-based allocation of test-time computation provides a favorable efficiency-effectiveness trade-off with a single efficient LLM.

Function renaming in stripped binaries can substantially assist reverse engineers by improving code readability, yet it is a challenging task. The difficulty stems from the need to accurately capture function semantics from low-level binary code across diverse instruction sets, architectures, and compiler optimizations, and to express these semantics in concise, human-readable names. Existing approaches either inadequately capture comprehensive function semantics or exhibit limited generalization to previously unseen binaries. In this paper, we present Hieronym, a generative large language model (LLM)-based framework for stripped binary function renaming. Hieronym adopts a hierarchical summarization-driven domain adaptation strategy and integrates multi-source information, including global binary context, local calling context, and intrinsic function semantics, to enhance the LLM's understanding of binary code. To enable systematic evaluation, we further propose a dual-layer evaluation framework that incorporates both token-level and whole-name-level metrics. We evaluate Hieronym on binary functions compiled with four compiler optimization levels (O0-O3) for four architectures (x64, x86, ARM, and MIPS). Experimental results demonstrate that Hieronym significantly outperforms state-of-the-art methods, achieving token-level improvements of 50.12% in precision, 41.75% in recall, and 45.10% in F1-score, as well as a 79.94% improvement in name-level accuracy, while also exhibiting strong generalization capability. Moreover, experiments on real-world malware samples further validate the practical effectiveness of Hieronym in security-critical scenarios.