生成模型与LLM推理优化

Scattered light is common in biomedical images, yet its removal remains challenging. The difficulty arises from three aspects: first, aligned scattered-light-free biomedical ground truth is often unavailable; second, scattering is coupled with weak illumination and sensor-induced noise; and third, many learning-based restoration models are computationally expensive for embedded devices in Internet of Medical Things (IoMT) scenarios. To address these issues, this paper proposes PP-Net, a hybrid physical-prior neural network for biomedical scattered light removal. The proposed method consists of three components: DFN-Net suppresses sensor-induced noise, ASAP estimates the scattering map and recovers a physics-based prior map, and GF-Net refines the prior map by fusing it with the denoised observation. To reduce the dependence on paired biomedical ground truth, a progressive synthetic training and cross-domain transfer strategy is developed. Experiments show that the physical-prior branch improves the peak signal-to-noise ratio (PSNR) by up to 1.26 dB on paired synthetic benchmarks. Under joint noise-and-scattering degradation, PP-Net improves PSNR by more than 10.8 dB and the structural similarity index measure (SSIM) by more than 0.62 compared with representative baseline methods. On real W2S biomedical images, the proposed method reduces the average Natural Image Quality Evaluator (NIQE) score by 43.3%. Edge deployment with RKNN conversion and INT8 quantization achieves an average inference latency of approximately 200 ms per $512\times512$ image over 360 test images. These results demonstrate that PP-Net provides an effective and deployable solution for microscopic imaging, endoscopic inspection, and edge-assisted biomedical analysis in IoMT scenarios.

Self-attention-based vision backbones perform well on dense prediction, but the quadratic computational cost of global softmax attention limits their efficiency as image resolution increases. We introduce Gated Token Recurrence (GTR), a softmax-free recurrent vision backbone that combines gated linear attention, alternating spatial scan directions, and spatially enhanced SwiGLU blocks. GTR is distilled from a detection-specialized DINOv3 teacher using only final-layer patch-token alignment through a linear projection and squared $\ell_2$ loss, without masked-token prediction or intermediate-layer supervision. With Objects365 detector pre-training, GTR-L achieves 58.9 box AP on COCO val2017 with 1.908\,ms median batch-one latency under compiled FP16 execution on an RTX~4090. The same backbone also transfers to instance segmentation, pose estimation, oriented detection, semantic segmentation, and monocular depth estimation. In an isolated kernel benchmark, our specialized chunkwise CUDA operator is $4.0\times$ faster than FLA v0.5.0 at 1.6K tokens on RTX~4090. TensorRT deployment on DRIVE AGX Thor achieves 2.282--8.769\,ms median batch-one latency across the evaluated models. These results show that recurrent token mixing can provide an efficient alternative to global softmax attention for high-resolution dense prediction and edge deployment. Project page: https://intellindust-ai-lab.github.io/projects/GTR/

Large language model (LLM) outputs are expected to be reproducible under greedy decoding, yet in practice the same model, prompt, and software stack produce different outputs on different GPUs. The root cause is floating-point non-associativity combined with hardware-dependent kernel selection. Inference frameworks select different matrix-multiplication kernels on each architecture, with different parallel reduction orders and unspecified tensor-core arithmetic, and the resulting rounding differences can flip output tokens. Existing solutions have imperfect cross-architecture reproducibility and incur a significant performance penalty. We present a solution employing a set of fixed-configuration fused-upcast GEMM kernels that load 16-bit weights from memory, upcast them to FP32 in registers, and accumulate with IEEE-754 arithmetic in a reduction order that is a pure function of the problem shape and is therefore independent of the device, its SM count, or kernel scheduling. By fixing the floating-point reduction order as a function of problem shape alone, every GPU runs the same operation sequence, so cross-architecture reproducibility of the linear layers reduces to correct IEEE-754 arithmetic rather than to rounding differences staying below a tie-flip threshold. We confirm our solution's linear-layer outputs are bitwise identical across NVIDIA Ampere, Ada, and Hopper GPUs, while running $1.17$ to $3.1\times$ faster end-to-end than the state-of-the-art solution and cutting weight-memory traffic in half.

Real-time avatar services run long-lived, GPU-backed sessions that transform a continuous text stream into speech, facial motion, rendered video, and RTC media. Elastic operation repeatedly reassigns these sessions across Workers. A reassignment updates shared placement state while the previous GPU runtime and RTC connection remain alive, opening a gap between control-plane assignment and execution authority. A stateful controller serializes ownership by retaining per-session authority, placing that state on the controller's scaling and failover path. We present DHSched, a control plane that manages long-lived stateful sessions through stateless peer Dispatch replicas. DHSched externalizes session ownership as a generation identified by (WorkerID, epoch). Dispatch conditionally commits each generation. Infer-Controller claims and revalidates the same generation before releasing state-advancing input. A stale runtime loses execution authority even if it remains alive. DHSched ranks candidates from periodically refreshed load views while keeping capacity admission authoritative. Recovery advances the same generation path via source-epoch guards and source-independent reconstruction. DHSched runs in our production avatar-serving stack, which reached 49,987 concurrent live sessions. On 10 June 2026, it completed 9,927 session migrations with no observed dual-owner conflicts or capacity overshoots. At the production peak, CreateLive latency was 98.3 ms at P99 and PushText latency was 16.5 ms at P99. Across 100,000 randomized ownership runs, assignment-time, claim-only, and WorkerID-only protocols permit stale execution, while DHSched records no dual-owner or stale-generation release.

Virtual try-on (VTON) requires precise pixel-level fidelity, yet mainstream Diffusion Transformers (DiTs) often suffer from texture degradation and structural drift. We identify symmetric interactions in standard joint-attention mechanisms as a source of these failures. Although such interactions support semantic flexibility in general-purpose editing, they allow stochastic noise to corrupt deterministic garment features in VTON. We analyze this problem through asymmetric information dynamics and introduce two diagnostic indicators: Conditional Attention Entropy (CAE) for feature unbiasedness and Injected Information Flux (IIF) for injection effectiveness. Our analysis suggests that symmetric bidirectional attention can corrupt conditional features and attenuate the conditional signal. To address these limitations, we propose RealFit, a framework that combines Unidirectional Information Flow (UIF) with Decoupled Timestep Modulation (DTM). UIF isolates the garment condition from stochastic noise to preserve garment identity, while DTM optimizes the modulation scale to maintain a strong conditional signal. The resulting time-invariant condition branch enables a conditional KV cache that reduces inference time by approximately 75%. RealFit offers a principled approach to conditional generation with state-of-the-art fidelity and efficiency.

Mixed-precision weight quantization is commonly formulated as a Multiple-Choice Knapsack Problem (MCKP), yet existing solvers rely on scalar sensitivity proxies that collapse each weight matrix's Hessian into a single number and treat every module independently. We prove that even the optimal scalar proxy incurs multiplicative distortion up to $\sqrt{κ(\mathbf{A})κ(\mathbf{B})}$ relative to the full activation-aware quadratic, where $κ(\mathbf{A})$ and $κ(\mathbf{B})$ denote the condition numbers of the input- and output-side Hessian factors. This bound varies from $10^1$ to $10^{13}$ for typical LLM modules, making inter-module sensitivity ranking unreliable. To address these limitations, we propose Cross-layer Activation-aware Sensitivity Allocation (CASA), a two-phase method. In Stage 1, the scalar proxy is replaced by an activation-aware metric derived from the Kronecker-factored Hessian, reducing the MCKP to a form whose continuous relaxation admits a closed-form solution. In Stage 2, a cross-layer-aware local search evaluates bit-width updates using the end-to-end model loss. Experiments on multiple LLMs across different bit budgets show that CASA achieves lower perplexity than the latest scalar-proxy baselines, especially at ultra-low bit-widths ($<3$ bits per weight). Moreover, the performance gain in zero-shot accuracy tracks the per-model average condition-number over modules, confirming the distortion bound as a practical indicator of scalar-proxy failure.

Quantization-aware distillation (QAD) restores much of the short-form question-answering performance lost to sub-3-bit quantization, yet leaves mathematical and code reasoning substantially impaired. Long generations often degenerate into repetitive loops, exhausting the decoding budget without completing a solution. We trace this gap to quantization-amplified exposure bias: QAD trains on fixed corpus prefixes, while quantization-induced deviations compound along the model's own autoregressive trajectories. To address this mismatch, we introduce an on-policy distillation (OPD) stage that places teacher supervision where the quantized model actually goes. Starting from a QAD checkpoint, the student generates through the quantized forward path used at deployment and receives feedback from a frozen full-precision teacher on its own prefixes, combining dense token-level guidance with task-verifier rewards. Across four models at 2.79 and 1.88 effective bits, OPD raises average BF16 performance retention from 35% to 70% on MATH-500 and from 66% to 91% on HumanEval while preserving short-form performance, with reasoning gains substantially exceeding those of continued teacher-forced QAD in matched-budget comparisons. By coupling QAD's stable low-bit initialization with OPD's on-policy reasoning recovery, our framework provides a comprehensive sub-3-bit solution that preserves broad capabilities while restoring long-form reasoning.

Attention computation makes inference expensive in video diffusion transformers (vDiTs), which generate videos through iterative denoising. Block-sparse attention (BSA) reduces this cost by computing only blocks selected by a logical mask, which specifies attention interactions to compute. However, coupling logical block geometry to execution choices limits adaptation to varying masks and graphics processing units (GPUs), while runtime kernel specialization can incur preparation overhead that outweighs execution time savings. We present Tessera, a specialized runtime for dynamic BSA that decouples logical masks from GPU execution while preserving specified attention interactions. Its physical mapping layer retains, combines, or subdivides logical attention blocks into physical tiles suited to different attention mask shapes and GPU architectures. Its task organization layer groups and schedules tiles within GPU tasks to reuse data, expose parallelism, and overlap data movement with computation. Finally, profile-guided regime selection enables low- overhead execution plan selection through a lookup table constructed from offline profiling. We implement Tessera with specialized CUDA kernels supporting four NVIDIA GPU generations. Evaluated on 2,315 real attention masks and industrial video diffusion models, Tessera achieves up to 6.79x BSA request speedup over baseline systems in the evaluated video diffusion models.

Large language model (LLM) serving is increasingly agentic, with multi-turn sessions that idle between actions yet must retain their full context. Limited GPU memory capacity forces inactive KV states to be evicted, so resuming a session incurs either costly recomputation or slow interconnect transfers. To address this, high bandwidth flash (HBF)-an on-package 3D-NAND memory offering orders-of-magnitude greater capacity than high bandwidth memory (HBM) at comparable read bandwidth-has emerged as a strong candidate. However, its high read energy and limited write endurance make it impractical to serve all KV traffic. Fortunately, our analysis shows that agentic KV states exhibit distinct access patterns: a small hot set is read for every decoding step, while a large cold pool is read only when a paused session resumes. Exploiting this, we place the hot set in HBM and the cold pool in HBF, forming a hot-cold KV hierarchy within the GPU memory tier. On agentic workloads with Qwen3-Coder-30B-A3B, our design delivers 14 ms time-between-tokens (TBT) and adds only $\approx$0.1 ms of resume latency on top of prefill, while hosting $24\times$ more concurrent sessions per GPU. By confining HBM to the hot set, our design also cuts read power by 7.6 kW per 8-GPU node relative to serving all KV from flash-establishing HBF as a cold-tier complement to HBM rather than its replacement.

We introduce SWE-Serve, a benchmark for evaluating agents on production inference engineering tasks. Implementing an inference feature can require coordinating multiple changes across the serving stack, including model support, runtime execution, and public APIs. Existing benchmarks provide limited coverage of production inference engineering: repository-level software engineering benchmarks do not target inference, while general terminal-agent benchmarks include only a few inference tasks. Dedicated inference benchmarks, meanwhile, focus primarily on isolated kernel generation or performance optimization rather than repository-scale production feature implementation. SWE-Serve provides 53 repository-grounded tasks derived from recent production changes to SGLang, spanning six inference engineering families. Each task executes on either CPU or a single GPU (H100) and is evaluated with hidden functional and regression tests, including, where applicable, end-to-end (E2E) serving tests and calibrated performance gates. Executable no-op and oracle controls, adversarial verifier review, and closed-book execution support task validity and evaluation integrity. Across 11 models and 31 model-effort configurations, the best-performing configuration achieves 75% mean pass@1. SWE-Serve exposes a substantial gap between completing tasks locally and achieving production correctness. On 19 tasks with end-to-end coverage, model-serving E2E tests reject roughly one-third of patches that pass every other test (45.9% under the verifier versus 69.4% with E2E tests excluded from scoring), with pass rate increasing for each model's best-performing configuration. By making the production correctness gap directly measurable, SWE-Serve enables the field to track whether future agents move beyond completing tasks locally to achieving production correctness.

Autoregressive OCR vision-language models accurately convert document images into text and structured markup, but require one sequential decoding step per output token, limiting inference speed. Unlike open-ended text generation, OCR outputs are strongly grounded in the input image, making diffusion-based parallel generation promising. However, when several tokens are predicted in one diffusion step, each is predicted before the others are known. Committing them directly can therefore introduce errors. We therefore introduce GravityOCR, a parameter-shared AR-block-diffusion model jointly trained for parallel drafting and causal AR verification. Verifying drafts before commitment lets the model commit multiple output tokens per round without a separate drafting network. The causal AR path also enables GRPO with sequence- and structure-level OCR rewards, avoiding diffusion-trajectory likelihood estimation while updating the shared drafter parameters. On OmniDocBench v1.6, AR-path GRPO improves the Overall score from 94.92 to 95.16 without reducing diffusion drafting efficiency, while the final model remains close to the original GLM-OCR score of 95.48. In an SGLang serving deployment, GravityOCR commits an average of 9.7 output tokens per forward pass and achieves a $3.94\times$ decode-only speedup on region crops and a $1.32\times$ end-to-end page-processing speedup over AR decoding.

Large language model (LLM) inference is constrained by the quadratic scaling of self-attention and the linear scaling of the KV cache, increasing latency, energy consumption, and GPU memory demand as context length scales. Existing soft-compression methods either lack query-guided memory selection at inference time, train without answer-targeted supervision, or couple compression tightly to a specific decoder architecture. We propose a Context-to-Answer-Aligned Memory Compression (CMC) framework, which compresses long input contexts into compact Context Memory Embeddings (CMEs) aligned to any frozen decoder's embedding space, reducing inference costs without modifying decoder weights. CMC introduces a two-tier KV cache that combines question-guided CME selection with a local context window, and trains the compressor with answer-targeted distillation from a frozen LLM. Experiments across nine encoder-decoder combinations and four QA benchmarks show that CMC consistently outperforms the baseline, achieving up to 7.3 EM and 4.0 F1 point gains on SQuAD, while reducing inference time and energy consumption by up to 20% and peak reserved GPU memory by up to 50% at 3,000 generation tokens. Ablation studies confirm that each architectural component and training objective contributes to the performance.

Despite their strong performance, large language models (LLMs) are bottlenecked by KV cache memory traffic during long-context inference. Sparse attention is widely used to accelerate LLM inference by computing exact attention over a selected subset of tokens. To recover the contribution of tokens excluded from exact attention, recent methods apply coarse-grained compensation to the omitted attention tail. However, existing methods typically select tokens based on attention mass and only then compensate for the unselected tokens. This decoupled design overlooks their interaction: selection should prioritize tokens that would leave the largest compensation error if omitted. To address this limitation, we introduce CompKV, the first compensation-aware sparse attention framework that divides tokens into blocks and explicitly optimizes selection for the downstream compensation mechanism. Our theoretical analysis shows that the residual left by block-level mean compensation is governed by both block attention mass and within-block logit variation. We approximate this residual using compact block-level statistics, yielding a deployable selection criterion. We further develop an efficient asynchronous implementation. Experiments on RULER and LongBench-Pro show that CompKV performs best among the evaluated sparse baselines while delivering up to a $6.85\times$ self-attention speedup over full attention.

Recent advances in large language models (LLMs) are driving the emergence of multi-modal and agentic services for mobile users through cloud and edge infrastructures, where long-context workloads pose daunting challenges for inference latency. Existing disaggregated LLM serving systems largely rely on hardware profiling, configuration enumeration, or heuristic scheduling, offering limited analytical guidance for cost-efficient resource allocation. In this paper, we propose SARA, a Service level objectives (SLOs)-Aware Resource Allocation framework for disaggregated agentic LLM serving systems, which maximizes goodput under a deployment cost constraint and a series of quantile-based SLO constraints. By capitalizing on queuing theory, we first model the prefill, KV cache transfer, and decode stages as an M/G/k queue, an M/G/1 queue, and a generalized birth-death process, respectively. The analysis reveals that the prefill and decode stages are dominantly limited by computational capacity and high-bandwidth memory (HBM) resources, respectively. With these mathematical models, we further derive tractable tail behaviors of different stage-wise service level metrics for both light- and heavy-tailed workloads. These characterizations explicitly map workload, model architecture, and hardware parameters to stage-wise SLO constraints and minimum resource requirements. Finally, we develop an effective resource allocation framework to maximize system goodput under limited cost budgets. Simulation and hardware results demonstrate that the proposed framework accurately predicts the stage-wise SLO with mean errors below 5%, and improves system goodput by 26.6% on average over state-of-the-art baseline methods under the same deployment cost.

Prefill and decode reward different approaches to quantization: low-precision arithmetic accelerates prompt processing, while compact weights reduce memory traffic during generation. We propose "disaggregated quantization" (DQ), which specializes computation formats, weights and storage placement to both of these phases. On Qwen 3 and Gemma 3, removing activation quantization specifically on decode improves accuracy on decode-heavy tasks without increasing inference cost. Training separate compute-native prefill weights accelerates prompt processing relative to weight-only inference while matching or exceeding its accuracy at 2-3-bit decode on both decode-heavy and prefill-heavy tasks. With released Qwen3.8-27B GGUF decoders, training an NVFP4 prefiller improves 1-bit accuracy by 32.5 points on MMLU-Pro and 35.3 on MMMU-Pro without modifying the decode checkpoint. To accommodate the additional checkpoint on a single device, offloaded disaggregated prefill (ODP) streams its weights from SSD, amortizing loading over prompt length. On the same 27B model, ODP delivers a 1.78x time-to-first-token speedup over the weight-only baseline at 8K prompt length in llama.cpp. We evaluate accuracy under disaggregated serving in vLLM and further validate shared-weight format disaggregation through post-training quantization on models up to 2.8T parameters.

Diffusion Large Language Models (dLLMs) have recently emerged as a promising alternative to autoregressive LLMs by enabling non-autoregressive text generation. However, their practical deployment remains limited by inefficient inference, largely due to the absence of effective Key-Value (KV) caching and scalable parallel decoding mechanisms. Existing acceleration methods typically study KV caching and parallel decoding in isolation, overlooking the I/O bottlenecks that arise when cache reuse and parallel token verification are jointly applied. In this work, we introduce $Flash-dLLM$, a training-free inference acceleration framework for fast and memory-efficient dLLMs. Flash-dLLM first identifies GPU memory I/O as a dominant bottleneck in KV-cache-enabled dLLM inference and addresses it with an I/O-aware fused KV-cache kernel that reduces redundant memory movement. Building on this optimized cache mechanism, Flash-dLLM further proposes an efficient KV-cache-driven draft-and-verify decoding strategy, where the dLLM itself serves as both drafter and verifier without requiring an auxiliary model. This unified design enables faster decoding while preserving generation quality and improving scalability to longer sequences and larger batch size. Extensive experiments on mathematical reasoning and code-generation benchmarks demonstrate that Flash-dLLM consistently outperforms existing state-of-the-art dLLM acceleration methods in both inference speed and memory efficiency. In particular, it achieves $5.1\times$ and $11.0\times$ speedups over prior strongest baseline Elastic-Cache on GSM8K and HumanEval, respectively.

Long-horizon and multi-turn agents typically generate short actions and process long observations from tools and environments. This growing context demands efficient prefill, compact KV-cache storage, and accurate long-context retrieval. To meet these demands, we introduce HySparse2, a hybrid sparse attention architecture with two-level KV sharing. At the outer level, KV Bridging adopts a YOCO-style self-decoder and cross-decoder structure, but bridges only full-attention layers. The self-decoder uses hybrid sliding-window attention (SWA), while the cross-decoder uses hybrid sparse attention. The KV caches for full-attention layers in the cross-decoder are generated from the hidden states of full-attention layers in the self-decoder. At the inner level, HySparse2 retains HySparse's core KV Reuse design with two refinements. First, it replaces block-level sparsity with token-level sparsity for finer long-context retrieval. Second, it removes the separate SWA branch from sparse layers and instead forces a sliding window of recent tokens into the sparse selection. This two-level KV sharing allows all cross-decoder KV caches to be constructed from self-decoder hidden states. Prefill can therefore exit after the self-decoder, skipping all cross-decoder layers. On an 80B-A3B MoE model, HySparse2 outperforms HySparse and Hybrid SWA on long-context retrieval and multi-turn agentic tasks, while substantially reducing prefill computation and KV-cache storage.

Video world models achieve long-range temporal consistency by storing KV cache during generation, but the growing cache makes KV cache memory a major deployment bottleneck, which motivates low-bit quantization study for efficiency. Existing 2-bit KV cache quantization methods can achieve nearly lossless performance on VBench, however, when applied to video world models, we find they still cause severe temporal flickering and visual degradation. Meanwhile, deeper investigates show that Key quantization produces smaller reconstruction errors than Value, but surprisingly leads to larger output degradation. We trace this discrepancy to attention in video world models: Key perturbations can change the attention logits, and shift the temporal-spatial tokens selected by Queries. These observations motivate us to preserve attention logits and temporal-spatial token selection during KV cache quantization. To address this issue, we present QuantWM, a training-free 2-bit KV cache quantization framework for video world models. QuantWM introduces two complementary techniques to mitigate the attention shifts. Firstly, quantization-sensitivity-aware clustering (QSAC) jointly considers historical Query sensitivity and residual ranges to select INT2-friendly Key centroids, which reduces quantization errors in channels that are more critical to attention. In addition, principal-subspace attention compensation (PSAC) restores the remaining Key errors along the dominant Query subspace using low-rank projections, which provides a direct and efficient correction to stabilize attention logits. Experiments on LingBot-World-v2, HY-World 1.5, Matrix-Game-2, Longcat-Video and Causal-Forcing demonstrate that QuantWM significantly improves visual quality and temporal consistency, while outperforming existing methods across benchmarks with up to 6.20 KV cache memory compression and limited additional overhead.

Greedy decoding from large language models is commonly treated as deterministic. We show it is not precision-invariant: the same model, prompt, and decoding algorithm produce different outputs in BF16 versus FP16 on identical hardware. Across our evaluations of six models (1.1B-7B parameters, four families; divergence additionally characterised at 12B) and three benchmarks, 49-100% of prompts diverge; a single token flip often cascades into trajectory-level divergence. We develop an empirical error-propagation analysis and find that 22 layers of accumulated body error do not distinguish flipping from non-flipping steps; the outcome depends primarily on the top-two logit margin at the LM head relative to the directional perturbation between the top-two candidates. The analysis makes five testable predictions about intervention outcomes, including that applying more FP32 compute (broader scope) makes agreement worse. The experiments match all five predictions. The best-performing low-overhead intervention we evaluate, selective FP32 LM head recomputation, triggered only when the margin falls below a threshold, delivers +22-36 pp exact agreement on A10G (+12-21 pp on L4 and A100) at less than 4% latency overhead in low-batch (batch size <=4) single-stream inference. We map the applicability boundary across six models and four batch sizes, and hypothesise that training-time precision stability is a determining factor. The method is a partial mitigation rather than a universal determinism guarantee: its benefit vanishes when body-originated error dominates, including at batch size >=8 and under end-to-end FP8 in our tests.

Training-free visual-token pruning often uses token importance, redundancy, or related selection criteria as proxies for safe removal. We show that these signals alone do not fully characterize removability, which is conditioned on both representation depth and the surrounding deletion set. Controlled interventions demonstrate that removing the same tokens at different depths produces substantially different downstream perturbations, while changing only the deletion context at a fixed depth alters candidate marginals and pruning-boundary decisions. These findings show that token importance alone cannot determine when a token is safely removable or how its removability changes under joint deletion. Motivated by this perspective, we propose CoRePrune, a training-free two-stage framework. Progressive Perturbation-Aware Visual Pruning refreshes deletion effects as visual representations evolve, while Set-Conditioned Refinement reevaluates candidate rescue benefits under the current deletion set after visual--text interaction. Across five multimodal large language model backbones covering standard images, high-resolution inputs, and video, CoRePrune preserves performance under aggressive token budgets. On Qwen3.5, with a final budget of 128 visual tokens, it retains 90.3% of dense-model performance while reducing aggregate prefill time by 51.0%.

Operational pseudospectral atmospheric models such as the ECMWF Integrated Forecasting System (IFS) run today almost exclusively on CPUs; GPU ports are under active development but not yet used in production. These models rely on the Spherical Harmonic Transform (SHT). Each time-step requires forward and inverse SHTs, and both passes depend on global pencil transposition that redistribute multi-dimensional arrays across compute nodes. At large node counts these global collectives dominate wall-clock time. We investigate GPU-resident lossy compression, using representative fields from the DYAMOND high-resolution operational dataset as input, and combining measured GPU compression throughput with SimGrid network simulation, we show that ZFP at 16 bits per value (rate-16), about the same storage budget as float16, matches the communication-time reduction of float16 truncation while delivering approximately $\mathbf{1600\times}$ lower mean relative error. ZFP at 8 bits per value (rate-8) achieves approximately $\mathbf{1.93\times}$ the speedup of float16 while retaining $\mathbf{4\times}$ lower mean relative error.

Emulation of high-accuracy floating-point matrix multiplication, as in the Ozaki scheme, splits the inputs into low-precision components and multiplies them pairwise. These products must be error-free, and each is an integer matrix product times a scale factor. FP4 Tensor Cores are the fastest on the NVIDIA B200 and B300 but cannot hold INT8 operands. The FP4 values scaled by 2 form the set $S = \{0, \pm1, \pm2, \pm3, \pm4, \pm6, \pm8, \pm12\}$, which contains every residue modulo 13, so with carries any integer splits into base-13 digits that FP4 can store. Prior work splits each INT8 operand into 3 such digits (limbs) with weights $(1, 13, 169)$ and multiplies them pairwise, 9 FP4 matrix multiplications (GEMMs) for INT8$\times$INT8. The classical ways to reduce products, such as the Karatsuba and Toom--Cook methods, do not apply as they stand: sums of limbs reach $\pm 24$ and leave $S$. This paper asks how many FP4 GEMMs are needed for one integer matrix product. We propose Adaptive Weight Encoding (AWE): the limbs take freely chosen integer weights, the stored planes are linear combinations of limbs, and the exact product is the sum of the FP4 GEMMs scaled by reconstruction coefficients. For each input range, we searched these choices for encodings with fewer products and found INT8$\times$INT8 in 6 products and INT4$\times$INT8 in 4. The formulation also holds modulo $m$, which covers the residue number systems of Ozaki scheme II: for the FP64 significand, the 75 products of prior work are reduced to 59. The boundary in product count between encodings with and without residues lies near input width 15. We release the encodings found.

Power availability increasingly constrains the operation of AI inference fleets, creating a need for provisioning methods that jointly consider serving capacity and power consumption. Prefill--decode (PD) disaggregation has emerged as a prevalent architecture for large-scale inference serving. However, determining the appropriate numbers of prefill and decode instances is challenging because serving capacity depends jointly on workload characteristics, hardware constraints, queueing, and KV-cache reservations. Existing approaches largely rely on profiling and simulation, providing limited analytical insight into how provisioning decisions shape the tradeoff between serving capacity and power consumption. This paper develops an analytical framework for power-aware provisioning of PD-disaggregated AI inference. Given an inference workload and hardware, the framework models the serving capacity and average power consumption of a provisioned deployment. The serving-capacity model is derived from the joint distribution of input--output lengths and hardware compute and memory limits. In particular, it explicitly captures the coupling between prefill and decode induced by KV-cache reservations, as well as the impact of request queueing. On this basis, the power model determines per-instance power consumption as a function of normalized serving throughput. Together, the models determine the serving capacity--power Pareto front among candidate provisioned deployments, enabling the service provider to choose a provisioned deployment as the workload or available power changes.

What limits KV-cache compression at extreme bit-rates? We argue that it is not the choice of compression scheme, but how its budget is allocated across attention heads. Existing methods apply rank and bit-width uniformly, ignoring that each head has a different optimal mix of rank truncation and quantization. We show that co-optimizing rank and bit-width per head, using only standard low-rank projection and scalar quantization, dominates uniform allocation, with the largest gains at low bit-rates. Our method, KV-COBRA (Co-Optimized Bit-Rank Allocation), formalizes this as a resource-allocation problem: it balances rank-truncation loss against quantization loss within each head, then redistributes budget across heads to minimize total distortion. A fused Hadamard rotation equalizes per-channel variance, and reordering the SVD basis by attention-KL importance makes the solver query-aware. The same allocator extends to joint $K{+}V$ compression. On perplexity, zero-shot, and long-context benchmarks from $0.5$ to $4$ bits per dimension (bpd), KV-COBRA shows the smallest accuracy degradation among evaluated methods at low bpd, with no per-token overhead.

Deploying Large Language Models for runtime operational triage incurs prohibitive latency (>100-500 ms), high VRAM requirements (>4-8 GB), and excessive energy dissipation. Extending Mandelbrot Fractal Neural Synthesis (Dagli et al., 2026), this paper presents the Universal Fractal Natural Language Decision Map, realized via the werr machine-native edge reflex runtime and the production answerr platform (https://answerr.me). Operating entirely without stored weight tensors (0 Bytes VRAM), the engine synthesizes deterministic decisions---noul (Boolean), choice (categorical), and score (ordinal)---by dynamically modulating 24-byte coordinate seeds along the chaotic boundary of the Mandelbrot set and evaluating multi-scale escape dynamics. Drawing inspiration from biological System-One reflex arcs, the engine introduces: (i) an Auto-Seed Router with domain projector Phi_D yielding a +28.8% accuracy gain over linear baselines; (ii) an Information-Theoretic Semantic Token Damping Filter (T_desc = 0.045) insulating against prompt injections (0.0% empirical bypass; 95% Wilson CI: [0.0%, 27.8%]) while pruning iterations by 45.8% (accelerating throughput 2.5x to 3.31 ms latency); (iii) a Multi-Scale Harmonic Tripod Fusion; (iv) a Coupled Margin Expansion Operator (Pitchfork Bifurcation Offset); and (v) a Cyclic Z/9Z Modular Resonant Grid Discretization based on the closed sub-ideal {0,3,6} (Lean 4 Mathlib ZMod 9), reducing FLOPs by 68.4%. Evaluated on JevBench (N=231), werr achieves 100.00% TypeSafe compliance and 81.65% calibrated accuracy with 7.08 ms median latency. We provide an OpenAI-compatible API and demonstrate deployment on 32-byte EVM smart contracts via the open-source werracle on-chain oracle (21,438 gas).

Modern information retrieval increasingly employs both embedding and generative models to handle complex queries. However, current serving systems suffer from low throughput and poor GPU utilization because they execute these models in isolation. Coarse-grained partitioning, such as dedicating GPUs to specific tasks, fails to adapt to dynamic workloads and creates computational "bubbles". To address these, we present Orthrus, a serving system that performs heterogeneous batching within a unified inference loop. The primary challenge lies in unifying embedding and generation workloads with conflicting computational patterns while optimizing batch composition for high performance. Orthrus addresses these challenges through chunked embedding with incremental pooling and by adjusting batch composition in a workload-aware manner. Evaluation on four A100 GPUs shows that, relative to baseline deployments, Orthrus achieves 1.28$\times$--4.52$\times$ higher throughput on controlled workloads and up to 55.8% lower end-to-end p99 latency on an iterative-RAG benchmark. We release our code at https://github.com/illinoisdata/Orthrus .

Large vision-language models (VLMs) can be efficiently deployed under stringent memory and latency constraints through post training quantization (PTQ). However, most PTQ methods are designed for unimodal large language models (LLMs). These methods treat quantization errors as isotropic perturbations under the Euclidean assumption, which provides weak guidance on directions most sensitive to quantization in VLMs. Consequently, directly adapting unimodal PTQ approaches or solely employing modality-specific scaling often leads to uneven bit-width distribution and inconsistent performance in low-bit settings. To address these challenges, we propose Riemannian Geometry-Sensitive Quantization (RGSQ), which formulates quantization as a reconstruction problem under a unified Fisher-Riemannian metric. RGSQ identifies modality-specific sensitive directions via Riemannian manifold mappings built from modality-partitioned empirical Fisher factors and fused into a modality-aware Kronecker-structured metric. We then apply geometry-aligned rotations to reorient the local tangent frame, steering low-bit perturbations toward loss-insensitive axes. Finally, we apply a whitening transformation that maps the Riemannian objective to an equivalent Euclidean form, enabling standard unimodal PTQ methods to evaluate multimodal quantization error under their original assumptions. Across an extensive and diverse set of mainstream VLM benchmarks, RGSQ achieves the highest accuracy and stability under extremely low-bit settings (W2A8 and W3A8). It outperforms VLM-aware baselines, such as MBQ and MQuant, by up to 5.9% and surpasses single-modality improvements by up to 8.6%.

Large language model (LLM) inference replicas run across tightly coupled GPUs and serve traffic continuously for weeks. Hardware and software failures are therefore inevitable, and one worker failure can disrupt an entire replica. Recovery requires reinitializing the engine, taking minutes even when weights and compilation artifacts are cached. Production deployments overprovision serving capacity to mask this window. We argue that the dominant cost is loss of ready serving capacity, not request progress, so recovery should preserve initialized engine state rather than reconstruct it. We present fast recovery for Dynamo based on this principle. Snapshots capture an initialized engine once and restore it instead of reinitializing it. Analysis of 18 weeks of failures from the Dynamo cluster shows that most failures are device-preserving: the engine process fails while the GPU and its resident allocations remain intact. Our key insight is that independent engine processes can reuse the same GPU-resident state while keeping mutable execution state private. The GPU Memory Service (GMS) decouples device-memory ownership from engine processes, enabling engines to share and reattach surviving allocations without copying them. GMS preserves model weights and shares them read-only between replacement and Shadow Engines, avoiding weight reloads. A second initialized runtime on the same GPUs reduces recovery to promotion. Across four models on vLLM and SGLang, these mechanisms recover a failed replica in under 7 seconds, 13-29 times faster than a warm restart, using a fixed 4-8 GiB of device memory per GPU independent of model size. Replaying the production trace, we estimate they would reclaim 79% of GPU-hours lost to recovery.

Latent diffusion models now dominate medical image generation, and every such pipeline rests on a tokenizer that compresses images into the latent codes for image generation to operate on. Thereby, the tokenizer choice bounds every downstream task from reconstruction fidelity and generation quality to the representations available for downstream analysis. Yet, medical imaging pipelines routinely utilize tokenizers from natural imaging on the hypothesis that their behavior carries over. However, this is an assumption never tested in the medical imaging regime, where datasets are orders of magnitude smaller and images exhibit far lower inter-sample variance. We present a systematic evaluation of medical image tokenizers evaluating thirty configurations across ten model families on twelve datasets at three compression factors, spanning reconstruction, generation, latent geometry, downstream classification, and memorization. We find that (1) performance on image reconstruction and generation strongly correlate, unlike prior reports on natural images; (2) modern tokenizers use nearly all of their codebook entries, but still leave most of the latent space unused; (3) training-set memorization is mild and is further suppressed by stronger latent space compression; and (4) discrete quantization can largely preserve downstream classification, with lookup-free schemes being the main exception.

Whole-slide pathology images (WSIs) contain gigapixel-scale visual content, creating a major scalability challenge for slide-level multimodal large language models (MLLMs). Existing approaches process thousands of patch tokens and typically apply compression only after slide encoding, leaving multimodal attention computationally expensive. We introduce SLICEChat, a slide-level MLLM that integrates progressive token pruning within a hybrid Mamba--Transformer slide encoder. Mamba layers enable efficient long-range propagation, while Transformer layers preserve global interactions as the sequence is progressively shortened. Between stages, language-supervised, region-aware pruning removes spatially coherent low-utility regions under a controlled keep-rate schedule, producing compact slide representations before multimodal fusion. On SlideBench VQA, SLICEChat achieves 79.84% accuracy on TCGA and 59.09% on BCNB cohorts, outperforming prior slide-level pathology MLLMs, and achieves the highest overall WSI-Bench metrics. It also provides competitive memory usage and the inference latency among the evaluated models. These results demonstrate accurate and computationally efficient multimodal reasoning over gigapixel WSIs.