Inference Latency

Large Language Models (LLMs) frequently exhibit "contextual disregard" when faced with input evidence that conflicts with their internal parametric memory, leading to persistent factual hallucinations. Existing mitigation strategies primarily rely on suppressing specific neuron activations or employing computationally expensive contrastive decoding mechanisms, which often result in increased perplexity or significantly elevated inference latency. To address these limitations, we propose Resonant Context Anchoring (RCA), a lightweight inference-time intervention method grounded in the perspective of residual stream signal dynamics. RCA aims to resolve the signal attenuation of external evidence during its propagation through deep networks. The core mechanism involves the orthogonal decoupling of routing logic and information magnitude within the self-attention module. By utilizing raw pre-softmax attention scores as an instantaneous metric of semantic alignment, we construct a dynamic gain field via non-linear rectification to selectively amplify the norms of value vectors corresponding to context tokens, without altering the attention probability distribution. This mechanism effectively elevates the signal-to-noise ratio (SNR) of input evidence within the residual stream mixture, thereby robustly anchoring the generation trajectory to the truthful context during inference. Extensive experiments on the Llama-3 model series demonstrate that RCA significantly improves contextual faithfulness across multiple factual consistency and strong knowledge-conflict tasks, effectively suppressing parametric hallucinations. Furthermore, results confirm that as a training-free and computationally negligible plug-and-play module, RCA achieves a Pareto improvement in faithfulness and fluency while maintaining the model's general language understanding capabilities.

Diffusion Large Language Models (DLMs) have demonstrated significant advantages across various tasks. However, constrained by their multi-step iterative inference mechanism, their computational overhead and inference latency in long-context tasks have become core bottlenecks restricting their large-scale deployment. When processing long sequences, existing Key-Value (KV) caching mechanisms often face a dilemma where generation quality degrades drastically, where the core challenge lies in precisely and efficiently filtering critical tokens within ultra-long contexts. Inspired by the human reading process, we propose \textbf{WaveFilter}, a universal and training-free caching framework. This framework innovatively introduces the wavelet transform for decomposition of long sequences to achieve precise identification of key tokens, based on which a sparse KV Cache is constructed to compute the final contextual representation. Experimental results demonstrate that WaveFilter, as a plug-and-play generic framework, significantly enhances the performance of existing mainstream KV Cache methods in complex long-context tasks.

Visual Autoregressive (VAR) models deliver high-quality image generation but suffer from significant inference latency at high resolutions. Recent acceleration approaches most rely on heuristic measures with layer features to prune tokens. Such heuristics are sensitive to complex contextual semantics, leading to inaccurate identification of redundant computation and poor adaptability across prompts. We rethink redundancy in VAR from the perspective of its impact on pixel-space generation and introduce Latent Discrepancy. This unified metric quantifies a token's contribution by measuring the change in model states during generation. Our analysis shows that redundancy is more accurately identified when guided by image latent or pixel-space signals. We further observed that in classifier-free guidance (CFG), the convergence trend of the discrepancy between conditional and unconditional branches exhibits high dynamics with different prompts. Based on these findings, we propose LD-Pruning (Latent Discrepancy Pruning), a training-free framework that removes redundancy via latent discrepancy by integrating decoding-free region selection and adaptive unconditional-branch skipping. Extensive experiments show that LD-Pruning substantially reduces inference latency while maintaining high generation quality, achieving up to 2.35x speedup on Infinity-8B.

Trajectory prediction is a fundamental task for autonomous systems, requiring complex reasoning about multi-agent interactions and intents. Large language models (LLMs) have recently been adopted for this task, as they provide strong contextual reasoning and interpretable, language-based trajectory representations. However, these LLM-based predictors are extremely memory- and compute-intensive, making them difficult to deploy on resource-constrained edge devices such as on-board computers in autonomous robots. To bridge this gap, we propose BitTP, which converts an LLM-based trajectory predictor into a lightweight bitlinear architecture. We demonstrate that weight-only quantization to 1.58-bit (BitTP-Weight) is optimal. Crucially, activations must remain in full precision, as quantizing them leads to severe degradation and instability in spatio-temporal reasoning. Empirically, BitTP-Weight not only preserves but improves prediction quality over the full-precision (BF16) LLM baseline, reducing ADE by 14.29% and FDE by 20.97% on average, while simultaneously reducing memory usage and inference latency relative to other quantization methods. These results demonstrate that carefully designed quantization acts as an effective regularizer, enabling the practical deployment of sophisticated LLM-based reasoning on edge devices. Code is available at: https://github.com/MintCat98/BitTP.

We study quality-latency-resource trade-offs in a documentation-grounded retrieval-augmented generation (RAG) system that uses Low-Rank Adaptation (LoRA) of the generator. We build a manually verified benchmark of 5,144 question-answer pairs over the official Kubernetes documentation and combine it with a fixed hybrid-retrieval pipeline (BGE-M3 dense, BGE-M3 native sparse, Reciprocal Rank Fusion, cross-encoder reranking). Over this benchmark we ablate 20 LoRA configurations on Llama-3.2-3B-Instruct and Llama-3.1-8B-Instruct across rank and target-module choices, and evaluate each on token-level F1, LLM-judged groundedness and correctness (pass@4), inference latency, inference memory, and training cost, all reported with bootstrap 95% confidence intervals. Pareto analysis shows that LoRA adapters acting only on the q and v attention projections consistently dominate the front, while the 3B/8B choice mainly defines operating regime. A param-matched control comparison further indicates that the q/v advantage is structural rather than purely parametric. The benchmark, selected adapters, and code are available at https://github.com/EugPal/rag-lora-tradeoffs.

Real-time semantic segmentation models offer an excellent balance between accuracy and inference speed. However, deploying these models in dynamic real world environments often requires the ability to learn novel classes incrementally without retraining on the entire dataset. This capability is known as continual learning. In this regard, the standard fine-tuning methods in deep learning often fail due to catastrophic forgetting, where the model learns new information but forgets previously trained and learned classes. Contributing to this crucial domain, the current paper proposes a novel continual learning framework tailored for PIDNet, which is a widely cited state-of-the-art real-time semantic segmentation model. Our method, PILOT(Parallel Incremental Learning Over Time), introduces a real-time and lightweight strategy by implementing a parallel Derivative-branch (D-branch) designed to capture the high frequency boundary information of novel classes while freezing the trained parameters of the original segmentation network. This novel setup allows the model to adapt to new semantic categories while preserving the knowledge of previously learned classes. By using only data associated with the new class, our model significantly reduces training overhead. Experimental results demonstrate that our approach successfully segments new classes while maintaining high mean Intersection over Union (mIoU) on the original base classes, thereby comfortably outperforming all major continual learning approaches in this domain. Overall, PILOT is shown to effectively mitigate catastrophic forgetting with minimal impact on inference latency, thus maintaining real-time performance.

Deep learning-based object detection has revolutionized Precision Livestock Farming (PLF), yet a critical barrier remains: high-performance Foundation Models (such as SAM 3) are too computationally intensive for edge deployment, while lightweight models (like YOLO) require prohibitive manual annotation efforts. This work proposes a fully automated knowledge distillation pipeline that leverages the Segment Anything Model 3 (SAM 3) to generate zero-shot pseudo-labels for training efficient YOLOv8 detectors. By treating SAM 3 as an offline auto-annotator, we eliminate the manual labeling bottleneck, producing models capable of real-time inference on resource-constrained hardware. We systematically evaluate this approach on the PigLife dataset, comparing SAM 3-supervised models against human-annotated baselines. Results demonstrate that a SAM 3-trained YOLOv8m achieves a mean Average Precision (mAP) of 79.4% without human intervention, while reducing inference latency by approximately 200$\times$ compared to the teacher model. Furthermore, stratified analysis reveals that in low-occlusion scenarios, the automated pipeline achieves detection rates comparable to human benchmarks ($AP_{50} > 99\%$). These findings indicate that foundation models can serve as effective, zero-annotation-cost supervisors, enabling scalable edge computing solutions for smart agriculture.

Time series forecasting drives operational decisions in areas like finance, transportation, and energy. While supervised learning approaches achieve strong performance, they require domain-specific training, feature engineering, and ongoing maintenance. Large-scale foundation models have recently emerged as a zero-shot alternative, avoiding task-specific training much like LLMs. In this work, we evaluate foundation models against standard supervised approaches. Rather than focusing solely on aggregate accuracy, we analyze performance across four operational regimes: periodic human-centric systems, physically constrained processes, stochastic financial markets, and heterogeneous demand forecasting. Our results characterize optimal deployment areas. Foundation models perform well in domains with transferable periodic structures and are efficient for cold-start or long-tail scenarios. Conversely, supervised specialists maintain higher precision in systems governed by strict physical constraints. In financial domains, newer foundation models are rapidly closing the performance gap with supervised specialists. We further quantify trade-offs in inference latency, data drift adaptability, and deployment constraints. Finally, we propose a Complexity Router that assigns each series to the optimal model class using empirical features. We demonstrate that this selective routing achieves higher accuracy and significantly lower inference costs compared to deploying a universal foundation model, providing a practical framework for balancing generalization and efficiency.

When Masked Diffusion Models (MDMs) generate sequences through iterative refinement, the rich internal computation over masked positions is discarded, forcing every subsequent refinement step to recompute the valuable internal information stored as model representations. To avoid a hard reset between denoising rounds, we propose Learned Relay Representations (Relay), a method that allows MDMs to be forward-thinking when denoising by explicitly learning how to propagate latent information for the benefit of future denoising steps. Relay introduces a differentiable per-token channel that passes information between forward passes and is trained via truncated backpropagation through time (BPTT). We show that this framework can be scaled to state-of-the-art Diffusion Language Models (DLMs), and is seamlessly compatible with techniques like block diffusion and KV caching. We first provide a thorough justification of the design choices in Relay on a challenging Sudoku-based planning task. We then scale Relay to Fast-dLLM v2, a state-of-the-art DLM, outperforming standard supervised finetuning on coding tasks while reducing inference latency by up to 32%. Our empirical results demonstrate that state-of-the-art DLMs can be explicitly trained to relay latent information forward across decoding steps, advancing the performance-latency Pareto frontier. We provide code for all our experiments.

Mobile imitation learning on portable demonstration interfaces faces two coupled bottlenecks: locomotion-contaminated action labels and inference-induced execution latency on a continuously moving base. Recent wrist-mounted interfaces lower the cost of tabletop data collection, yet a single wrist view does not capture the global context required for base navigation. Adding a body-mounted camera entangles human walking with hand motion. Meanwhile, generative policies introduce hundreds of milliseconds of inference latency, during which the base advances past predicted waypoints, forcing backward corrections at action splices. This paper presents Mobile UMI, a hardware-free demonstration framework that addresses both gaps through three components. First, a dual-camera capture system records chest-centric global context and wrist-centric local interaction without any robot present. Second, a one-shot ChArUco-based spatial anchor unifies the chest and hand visual-inertial frames; the hand pose is then re-expressed relative to the chest to extract decoupled SE(3) manipulation and SE(2) base trajectories. Third, an asynchronous receding-horizon executor performs online state matching: each generated action chunk is realigned with the current physical pose so that expired waypoints are discarded before execution. The full system is evaluated on four long-horizon household tasks, achieving an average success rate of 83.8% over 100 trials per task. Controlled comparisons against ACT and Diffusion Policy show that the chest-relative label alone closes much of the gap; online state matching closes the remainder. These results indicate that, for mobile imitation learning under the tested conditions, explicit kinematic factorization combined with state-level latency alignment provides an effective solution without requiring architectural changes to the underlying policy class.

In-context learning (ICL) allows large language models (LLMs) to adapt to new tasks through demonstrations, yet it suffers from escalating inference costs as context length increases. While task vectors offer a promising alternative by compressing demonstrations into compact hidden-state representations, their quality has been evaluated only through downstream task accuracy. This indirect criterion provides limited insight into how to design more effective task vector extraction methods. In this paper, we posit that inference using task vectors should align their predictive distribution with that of ICL. To quantify this, we introduce $d_{\text{NTP}}$, a metric that measures the discrepancy in next-token probabilities between task vector-based and ICL-based inference. Our empirical analysis reveals that $d_{\text{NTP}}$ serves as a performance proxy, exhibiting a strong negative correlation with downstream accuracy. Motivated by this, we develop Linear Task Vector (LTV), a method designed to minimize $d_{\text{NTP}}$ via a closed-form linear mapping that estimates demonstration effects through regression. Across eight classification benchmarks and five LLMs, LTV consistently outperforms existing task vector baselines, improving average accuracy by 9.2% while reducing inference latency. We further show that LTV outperforms the baselines on regression tasks. Moreover, we investigate the transferability of LTV across different model scales; an aspect that has remained nascent in task vector research. Specifically, we empirically show that task vectors from a larger model can enhance a smaller model's performance by 6.4%, suggesting a new utility for extracted task representations.

The Mixture-of-Experts (MoE) architecture improves computational efficiency via sparse expert activation, but throughput-oriented inference faces substantial GPU memory pressure due to a significant parameter size and intermediate data. Prior works attempt to mitigate this using expert offloading with micro-batching or by offloading computation to the CPU. However, the fragmented workload resulting from micro-batching degrades operational intensity, causing expert execution to become memory-bound. Meanwhile, CPU offloading is constrained by slow PCIe transfers and its limited applicability to attention computation in the decode stage. Consequently, these inefficiencies prevent effective system utilization, severely restricting the end-to-end throughput of MoE inference. To address these challenges, this paper proposes CoX-MoE, an Advanced Matrix Extensions (AMX)-enabled CPU-GPU collaborative system that comprehensively optimizes MoE inference by combining coalesced expert execution with strategic workload orchestration for higher throughput. CoX-MoE introduces (i) a coalescing-aware orchestration policy to jointly optimize resource allocation by adopting ordinary batch, instead of micro-batch, for expert computation and selective attention offloading, and (ii) a static expert-aware stratification scheme that pre-assigns frequently activated experts to the GPU, mitigating PCIe transfer overhead and balancing workload for the CPU and GPU during inference. Compared to state-of-the-art frameworks, CoX-MoE delivers significant gains, achieving up to 7.1x and 2.4x higher throughput than FlexGen and MoE-Lightning, respectively.

Large language models (LLMs) are increasingly needed for interactive mobile applications, but high-quality models exceed the limited DRAM available on smartphones. Flash storage can hold larger models, yet flash-backed inference is slow because autoregressive decoding repeatedly invokes the target model and incurs costly I/O. We observe that speculative decoding is a natural fit for this setting: a small draft model can remain in DRAM, while a larger flash-resident target model verifies multiple candidate tokens per invocation. However, existing methods assume server-class accelerators and fail to account for prolonged I/O latency, limited computation parallelism, and irregular speculation execution. We present Lever, an end-to-end system for efficient flash-backed LLM inference on smartphones. Lever jointly optimizes the three stages of speculative decoding under mobile constraints. For drafting, it builds token trees using an I/O- and compute-aware gain-cost objective. For verification, it prunes low-value branches through early-exit prediction to reduce target-model computation. For execution, it maps speculation efficiently across mobile CPU-NPU hardware to improve utilization. Comprehensive evaluations show that Lever reduces inference latency by an average of 2.93x over baseline flash-offloaded inference and 1.50x over conventional speculative decoding, narrowing the latency gap between flash-backed and memory-resident LLM inference.

Large language model (LLM) based multi-turn dialogue systems often struggle to track dependencies across non-adjacent turns, undermining both consistency and scalability. As conversations lengthen, essential information becomes sparse and is buried in irrelevant context, while processing the entire dialogue history incurs severe efficiency bottlenecks. Existing solutions either rely on high latency external memory or lose fine-grained details through iterative summarization. In this paper, we propose Self-Recall Thinking (SRT), a framework designed to address long-range contextual dependency and sparse informative signals in multi-turn dialogue. SRT identifies helpful historical turns and uses them to generate contextually appropriate responses, enabling the model to selectively recall and reason over context during inference. This process yields an endogenous reasoning process that integrates interpretable recall steps without external modules. SRT incorporates: (1) Dependency Construction: Generating and converting it into self-recall chains; (2)Capability Initialization: Training to enable reasoning chains with recall tokens capability; (3)Reasoning Improvement: Refining accuracy via verifiable rewards to optimize recall and reasoning for correct answers. Experiments on multiple datasets demonstrate that SRT improves F1 score by 4.7% and reduces end-to-end latency by 14.7% over prior methods, achieving a balance between reasoning latency and accuracy, and outperforming state-of-the-art baselines.

Diffusion-based vision-language-action models (dVLAs) are promising for embodied intelligence but are fundamentally limited in real-time deployment by the high latency of full inference. We propose Realtime-VLA FLASH, a speculative inference framework that eliminates most full inference calls during replanning by introducing a lightweight draft model with parallel verification via the main model's Action Expert and a phase-aware fallback mechanism that reverts to the full inference pipeline when needed. This design enables low-latency, high-frequency replanning without sacrificing reliability. Experiments show that on LIBERO, FLASH largely preserves task performance by replacing many 58.0 ms full-inference rounds with speculative rounds as fast as 7.8 ms, lowering task-level average inference latency to 19.1 ms (3.04x speedup). We additionally demonstrate effectiveness on real-world conveyor-belt sorting, highlighting its practical impact for latency-critical embodied tasks.

Large Reasoning Models (LRMs) are increasingly integrated into systems requiring reliable multi-step inference, yet this growing dependence exposes new vulnerabilities related to computational availability. In particular, LRMs exhibit a tendency to "overthink", producing excessively long and redundant reasoning traces, when confronted with incomplete or logically inconsistent inputs. This behavior significantly increases inference latency and energy consumption, forming a potential vector for denial-of-service (DoS) style resource exhaustion. In this work, we investigate this attack surface and propose an automated black-box framework that induces overthinking in LRMs by systematically perturbing the logical structure of input problems. Our method employs a hierarchical genetic algorithm (HGA) operating on structured problem decompositions, and optimizes a composite fitness function designed to maximize both response length and reflective overthinking markers. Across four state-of-the-art reasoning models, the proposed method substantially amplifies output length, achieving up to a 26.1x increase on the MATH benchmark and consistently outperforming benign and manually crafted missing-premise baselines. We further demonstrate strong transferability, showing that adversarial inputs evolved using a small proxy model retain high effectiveness against large commercial LRMs. These findings highlight overthinking as a shared and exploitable vulnerability in modern reasoning systems, underscoring the need for more robust defenses.

While the overall inference latency of Video Diffusion Transformers (DiTs) can be substantially reduced through model distillation, per-step inference latency remains a critical bottleneck. Existing acceleration paradigms primarily exploit redundancy across the denoising trajectory; however, we identify a limitation where these step-wise strategies encounter diminishing returns in few-step regimes. In such scenarios, the scarcity of temporal states prevents effective feature reuse or predictive modeling, creating a formidable barrier to further acceleration. To overcome this, we propose Frame Interleaved Sparsity DiT (FIS-DiT), a training-free and operator-agnostic framework that shifts the optimization focus from the temporal trajectory to the latent frame dimension. Our approach is motivated by an intrinsic duality within this dimension: the existence of frame-wise sparsity that permits reduced computation, coupled with a structural consistency where each frame position remains equally vital to the global spatiotemporal context. Leveraging this insight, we implement Frame Interleaved Sparsity (FIS) as an execution strategy that manipulates frame subsets across the model hierarchy, refreshing all latent positions without requiring full-scale block computation. Empirical evaluations on Wan 2.2 and HunyuanVideo 1.5 demonstrate that FIS-DiT consistently achieves 2.11--2.41$\times$ speedup with negligible degradation across VBench-Q and CLIP metrics, providing a scalable and robust pathway toward real-time high-definition video generation.

Speculative decoding speeds up autoregressive generation in Large Language Models (LLMs) through a two-step procedure, where a lightweight draft model proposes tokens which the target model then verifies in a single forward pass. Although the drafter network is small in modern architectures, its LM-head still performs projection to a large vocabulary, becoming one of the major computational bottlenecks. In prior work this issue has been predominantly addressed via static or dynamic vocabulary truncation. Yet mitigating the bottleneck, these methods bring in extra complexity, such as special vocabulary curation, sophisticated inference-time logic or modifications of the training setup. In this paper, we propose SlimSpec, a low-rank parameterization of the drafter's LM-head that compresses the inner representation rather than the output, preserving full vocabulary support. We evaluate our method with EAGLE-3 drafter across three target models and diverse benchmarks in both latency- and throughput-bound inference regimes. SlimSpec achieves $4\text{-}5\times$ acceleration over the standard LM-head architecture while maintaining a competitive acceptance length, surpassing existing methods by up to $8\text{-}9\%$ of the end-to-end speedup. Our method requires minimal adjustments of training and inference pipelines. Combined with the aforementioned speedup improvements, it makes SlimSpec a strong alternative across wide variety of draft LM-head architectures.

Modern Large Foundation Model (LFM) training has transformed the data pipeline from a static ingestion layer into a dynamic component that must co-evolve with the training process. Existing systems are ill-equipped: colocated dataloaders offer no failure isolation, while message queue-based disaggregated dataloaders operate on a record/offset abstraction that cannot express the batch-level semantics required by distributed training. We present BatchWeave, an object-store-native training data plane for distributed LFM training. BatchWeave uses versioned manifests and conditional object writes to coordinate batch publication, recovery, and lifecycle management. First, it introduces the Transactional Global Batch (TGB), which builds on versioned-manifest ACID storage semantics and extends them with training-specific consistency, including atomic all-rank batch visibility, a globally ordered step sequence, checkpoint-aligned lifecycle management, and end-to-end exactly-once recovery. Second, it realizes recovery and retention directly in the storage layer, by durably persisting producer state through the commit protocol and tying reclamation to distributed checkpoint state. Third, its Decentralized Adaptive Commit (DAC) algorithm sustains stable ingestion throughput as the manifest grows, without any inter-producer communication. Evaluations on large-scale multimodal pre-training and SFT workloads using 64 GPUs show that BatchWeave outperforms colocated dataloader throughput while providing full failure isolation, outperforms Apache Kafka in ingestion throughput, and achieves lower consumer read latency than Kafka.

Large-scale 3D vision-language models (VLMs) like LLaVA-3D offer strong spatial reasoning but are difficult to deploy due to high computational costs. We propose a knowledge distillation framework that transfers spatial reasoning from a 7B teacher to a 2.29B student model. Our approach achieves 8.7x lower inference latency and a 3x reduction in model size while retaining 54-72% of the teacher's performance. The framework utilizes VGGT as the vision encoder and a multi-task distillation pipeline with uncertainty-aware loss weighting. To improve reasoning without chain-of-thought (CoT) data, we introduce "Hidden CoT": learnable latent tokens that serve as an internal scratchpad before answer generation. This is the first use of latent scratchpad reasoning in distilled 3D VLMs. The student model jointly performs spatial description, depth estimation, and object detection. Experiments on ScanNet and 3D-FRONT show strong spatial understanding, reaching 68-72% accuracy on proximity and contact tasks. Our framework enables efficient 3D scene QA on resource-constrained platforms.

Mixture-of-Experts (MoE) models achieve remarkable performance by sparsely activating specialized experts, yet their massive parameters in experts pose significant challenges for deployment. While low-rank quantization offers a promising route to compress MoE models, existing methods still incur nonnegligible memory overhead and inference latency. To address these limitations, we propose \textsc{TileQ}, a fine-tuning-free post-training quantization (PTQ) method that employs 2D-tiling structured low-rank quantization to share low-rank factors across both input and output dimensions of MoE experts. Furthermore, we introduce an efficient inference technique for \textsc{TileQ} that fuses multiple low-rank expert computations into a single-pass operation, significantly improving hardware utilization. Experiments show that \textsc{TileQ} cuts down additional memory usage up to 10$\times$ and reduces inference latency to $\sim$5% while preserving state-of-the-art accuracy.

Chain-of-Thought (CoT) prompting symbolized a huge improvement of reasoning capabilities of Large Language Models (LLMs). However, scaling up test-time computation yields extensive CoT sequences, introducing severe inference latency and key-value (KV) cache memory bottlenecks. While structural pruning offers a fundamental, hardware-aware solution to alleviate static parameter burdens, existing magnitude-based methods may cut off the neurons of CoT: by over-indexing on discrete cross-entropy objectives, these heuristics fall into a \textit{magnitude trap}: they prioritize high-frequency, low-information syntactic tokens and trigger a disappointing reasoning collapse at high sparsities (e.g., 40%). To overcome this topological phase transition, we propose \textsc{Relative Kinetic Utility} (RKU), a novel theoretical framework that elevates discrete pruning to a continuous kinetic integral over the depth manifold of the model based on Alternating Gradient Flow(AGF). By modifying it with Fisher trace normalization, RKU acts as a lightweight curvature-aware normalization to isolate \textit{kinetic spikes} -- the fundamental structural pathways responsible for high-curvature logical routing. Extensive experiments on Qwen-2.5-7B and LLaMA-3-8B improves performance in the high-sparsity regime around 40%. RKU attains 13.34% accuracy on GSM8K at 40% sparsity, outperforming the strongest baseline, and appears to better preserve reasoning-relevant representations under out-of-distribution evaluation.

Reasoning-based end-to-end (E2E) autonomous driving has recently emerged as a promising approach to improving the interpretability of driving decisions as it can generate human-readable reasoning together with predicted trajectories. Such approaches commonly generate multiple trajectories to capture diverse future behaviors, and they fall into two categories: (1) multi-reasoning, where one reasoning sequence is generated per trajectory, and (2) single-reasoning, where a single reasoning is shared across all trajectories. The former offers richer diversity at the cost of redundant computation, while the latter is more efficient but is often assumed to sacrifice diversity. Alpamayo 1, a representative system, adopts the multi-reasoning approach and achieves competitive trajectory prediction performance. However, the efficiency of this design remains largely unexplored, making it a well-motivated subject for investigation. In this paper, we systematically analyze and improve Alpamayo 1 in two ways. First, we reduce inference latency while preserving trajectory diversity by redesigning Alpamayo 1 into a single-reasoning system. Through extensive experiments, we find that replacing multi-reasoning with single-reasoning does not meaningfully degrade trajectory diversity. Second, we accelerate diffusion-based action generation by eliminating inter-block overhead arising from unnecessary copy operations and inefficient kernel execution. Through closed-loop and open-loop experiments, we validate both optimizations, demonstrating a 69.23% reduction in inference latency while maintaining trajectory diversity and prediction quality. These results highlight the importance of jointly analyzing system architecture and runtime execution to improve the efficiency of reasoning-based E2E AD systems.

Over the last two decades, language modeling has experienced a shift from the use of predominantly recurrent architectures that process tokens sequentially during training and inference to non-recurrent models that process sequence elements in parallel during training, which results in greater training efficiency and stability at the expense of lower inference throughput. Here we introduce the Structured Recurrent Mixer, an architecture that allows for algebraic conversion between a sequence parallel representation at train time and a recurrent representation at inference, notably without the need for specialized kernels or device-specific memory management. We show experimentally that this dual representation allows for greater training efficiency, higher input information capacity, and larger inference throughput and concurrency when compared to other linear complexity models. We postulate that recurrent models are poorly suited to extended sequence length scaling for information-rich inputs typical of language, but are well suited to scaling in the sample (batch) dimension due to their constant memory per sample. We provide Mojo/MAX inference implementations of SRMs exhibiting 12x the throughput and 170x the concurrency of similarly powerful Transformers inferenced on vLLM, increases characteristic of Pytorch implementations resulting in a 30% increase in compute-constant GSM8k Pass@k. We conclude by demonstrating that SRMs are effective reinforcement learning training candidates.

Diffusion Language Models (dLLMs) have garnered significant attention for their potential in highly parallel processing. The parallel capabilities of existing dLLMs stem from the assumption of conditional independence at high confidence levels, which ensures negligible discrepancy between the marginal and joint distributions. However, the stringent confidence thresholds required to preserve accuracy severely constrain the scalability of parallelism. Through systematic token-level statistical analysis, we reveal that a substantial proportion of tokens converge to their correct predictions early in the denoising process yet fail to reach standard confidence thresholds, confirming that current confidence-based criteria are overly conservative. In response, we introduce LEAP (Lookahead Early-Convergence Token Detection for Accelerated Parallel Decoding). LEAP is a training-free, plug-and-play method that leverages future context filtering and multi-sequence superposition to detect early-converging tokens. By validating the alignment between early convergence and correctness, we enable reliable early decoding of these tokens. Benchmarking across diverse domains demonstrates that LEAP significantly lowers inference latency and decoding steps. Compared to confidence-based decoding, the average number of denoising steps is reduced by about 30%. On the GSM8K dataset, combining LEAP with dParallel accelerates decoding to 7.2 tokens per step while preserving model precision. LEAP effectively breaks the reliance on high-confidence priors, offering a novel paradigm for parallel decoding.

Diffusion models for multi-agent trajectory prediction are limited by iterative denoising, which causes inference latency that hinders their use in time-critical settings like autonomous driving. Fast-sampling variants using DDIM and informed initial noise distributions partially alleviate this issue, but they either fail to achieve true single-step generation or are constrained by the chosen noise distribution. Consistency Models (CMs) offer high-quality one-step generation by mapping noise directly to data, but are difficult to train from scratch. We propose ECTraj, an enhanced CM pipeline with improved training and conditional generation for trajectory prediction. Our framework extends the student-teacher consistency training scheme: the student produces standard outputs, while the teacher explicitly fuses its predictions with parts of the ground truth to give stronger supervision. We also exploit CMs' direct denoising for top-K multi-shot generation during training. Combining conditional generation with this enhanced consistency objective yields faster inference and improved prediction accuracy, establishing competitive new benchmarks on the large-scale Argoverse 2 dataset.

As the widespread adoption of Large Language Models (LLMs) accelerates, token consumption from intermediate reasoning traces increasingly contributes to inference latency and operational cost. Recent studies suggest that many real-world tasks require little to no explicit reasoning, with additional reasoning sometimes even degrading performance. In this work, we propose Post-Reasoning, a simple yet effective approach that improves instruction-tuned models by conditioning them to justify their answers after generating the final response. By design, it enables the final answer to be obtained without additional latency or token cost, while still improving performance through simple instruction augmentation. We evaluate Post-Reasoning across 117 model--benchmark settings spanning 13 open and proprietary models, 4 model families, and 9 diverse reasoning and knowledge-intensive benchmarks, including AMC, HMMT, GSM8K, GPQA, MMLU-Pro, and BIG-Bench Hard. Post-Reasoning improves performance in over 88.19% of evaluated settings, achieving a mean relative improvements of 17.37%. Furthermore, we propose supervised post-reason tuning, which further improves performance in over 91.11% of evaluated settings, and exceeds the prompt-based post-reasoning baseline by an average of 8.01%, demonstrating that post-reasoning can be effectively internalized through training. Ultimately, Post-Reasoning establishes a new performance ceiling for direct-answer capabilities.

Generative AI models, such as Large Language Models (LLMs) and diffusion models, have demonstrated impressive performance across a wide range of tasks. Despite these advances, deployment remains challenging due to substantial memory requirements, extended inference latency, significant computational demands, and high hardware costs. These issues are further complicated when evaluating models across heterogeneous platforms, where differences in numerical formats, memory bandwidths, and software stacks interact with model architecture and workload characteristics in complex ways. To address these challenges, we present a systematic study focused on performance optimization and comparative analysis of several Generative AI models across diverse downstream tasks. This work introduces a novel mixed-precision post-training quantization evaluation, examines fine-tuning strategies, and assesses performance across modern high-performance computing (HPC) systems and advanced accelerators.

Vision-Language-Action (VLA) models offer a promising path to generalist robot control, but their inference latency causes observation staleness when generated actions are executed asynchronously. Several methods have been proposed concurrently to mitigate this problem: inference-time inpainting (IT-RTC), training-time delay simulation (TT-RTC), future-state-aware conditioning (VLASH), and lightweight residual correction (A2C2). Each takes a fundamentally different approach, but they have so far been evaluated independently with different codebases, base policies, and protocols. We present a systematic comparison of these four methods under controlled conditions. We develop two unified codebases that integrate all methods with harmonized library and dataset versions, and we benchmark them on the Kinetix suite with MLPMixer policies and on the LIBERO manipulation benchmark with SmolVLA, sweeping inference delays up to $d=20$ control steps. A2C2's per-step residual correction is the most effective method on Kinetix, holding above 90% solve rate up to $d=8$, and also leads on LIBERO from $d=4$ onwards. IT-RTC is competitive at low delays but degrades sharply under long chunks ($H=30$) and high delays. TT-RTC is the most robust training-based method: stable across d_\max choices, generalizes beyond its training delay distribution, and adds zero inference overhead. VLASH exhibits a clear low-delay vs. high-delay trade-off governed by the fine-tuning delay range [0,d_\max]. Code is available at https://github.com/TheAyos/async-vla-inference

Recently, video language models (VLMs) have been applied in various fields. However, the visual token sequence of the VLM is too long, which may cause intolerant inference latency and GPU memory usage. Existing methods propose mixed-precision quantization to the key-value (KV) cache in VLMs based on token granularity, which is time-consuming in the search process and hardware inefficient during computation. This paper introduces a novel approach called WindowQuant, which employs window-adaptive mixed-precision quantization to optimize the KV cache. WindowQuant consists of two modules: window-level quantization search and window-level KV cache computation. Window-level quantization search quickly determines the optimal bit-width configuration of the KV cache windows based on the similarity scores between the corresponding visual token windows and the text prompt, maintaining the model accuracy. Furthermore, window-level KV cache computation reorders the KV cache windows before quantization, avoiding the hardware inefficiency caused by mixed-precision quantization in inference computation. Extensive experiments demonstrate that WindowQuant outperforms state-of-the-art VLM models and KV cache quantization methods on various datasets.

Weight pruning is widely advocated for deploying Large Language Models on resource-constrained IoT and edge devices, yet its impact on model fairness remains poorly understood. We conduct a controlled empirical study of three instruction-tuned models (Gemma-2-9b-it, Mistral-7B-Instruct-v0.3, Phi-3.5-mini-instruct) across three pruning methods (Random, Magnitude, Wanda) at four sparsity levels (10-70%) on 12,148 BBQ bias benchmark items with 5 random seeds, totaling 2,368,860 inference records. Our results reveal a Smart Pruning Paradox: activation-aware pruning (Wanda) preserves perplexity nearly perfectly (just 3.5% increase at 50% sparsity for Mistral-7B), yet produces the highest bias amplification, with Stereotype Reliance Score increasing 83.7% and 47-59% of previously unbiased items developing new stereotypical behaviors at 70% sparsity. Random pruning destroys language capability entirely (perplexity exceeding $10^4$ and reaching $10^8$) but produces only random-chance bias. We further show that unstructured pruning provides zero storage savings and zero inference latency reduction on real edge hardware, undermining the primary motivation for its use in IoT deployment. Of 180 dense-vs-pruned comparisons, 141 (78.3%) are significant ($p < 0.05$) with mean $|h| = 0.305$. Published quantization studies report up to 21% of responses flipping between biased and unbiased states; our pruning results show transition rates nearly three times higher (47-59%), suggesting pruning poses a categorically greater risk to alignment than quantization. These findings demonstrate that perplexity-based evaluation provides false assurance of behavioral equivalence, and that IoT deployment pipelines require bias-aware validation before deploying pruned models at the edge.

Neuron pruning is widely used to reduce the computational cost and parameter footprint of large language models, yet it remains unclear whether neurons in task-specific models contribute uniformly to task performance. In this work, we provide empirical evidence for the existence and importance of task-specific neurons through a systematic pruning study on language models specialized for mathematical reasoning and code generation. We introduce an activation-based selectivity metric to identify neurons with low contribution to the target task and prune them while preserving target-task accuracy, and compare selective pruning with random pruning. Selective pruning consistently outperforms random pruning, indicating that activation-based selectivity provides a systematic advantage over random pruning. Reverse pruning experiments further show that removing a small subset of highly task-specific neurons (~10%) causes complete performance collapse, suggesting that there exist task specific neurons and critical task information is concentrated in a small portion of the network. In contrast, selective pruning of less critical neurons (~30% - ~35%) reduces accuracy but still preserves significant performance. We also observed consistent reductions in parameters and runtime VRAM usage, along with improved inference throughput as pruning increases. Experiments on both 1.5B and 7B models reveal a robustness threshold around 15-20% pruning, beyond which accuracy loss and generation failures increase sharply. Fine-tuning substantially recovers performance across pruning levels, particularly for aggressively pruned models. These findings provide empirical evidence of neuron specialization in task-specific language models and offer insights into pruning robustness, model redundancy, and post-pruning recoverability.

Achieving a desirable balance between rendering quality and real-time performance is a long-standing challenge in modern game and rendering engines, particularly on resource-constrained mobile devices such as laptops, tablets, and smartphones. Existing approaches to automatic rendering parameter optimization either depend on exhaustive per-scene pre-computation that spans several days, suffer from the prohibitive inference overhead of neural networks that prevents per-frame adaptation, or lack generalizability across heterogeneous hardware and diverse scenes. In this paper, we propose \textbf{LUT-Opt}, a lightweight, general-purpose framework for adaptive per-frame rendering parameter optimization. Our method decomposes the joint optimization of rendering time and image quality into a tractable two-stage pipeline. In the offline stage, we train a pair of XGBoost regressors to predict rendering time and image quality from rendering parameters, hardware state, and scene complexity descriptors. The trained ensemble models are then distilled into compact lookup tables (LUTs) through systematic discretization and a two-phase linear search that first constrains rendering time and subsequently maximizes structural similarity (SSIM). During runtime, the pre-computed LUT is queried every frame in sub-millisecond time, enabling truly adaptive parameter selection with negligible computational overhead. We validate LUT-Opt on two representative rendering techniques -- subsurface scattering (SSS) and hybrid-pipeline ambient occlusion (AO) -- implemented within Unreal Engine 5. Extensive experiments across multiple scenes and GPU configurations demonstrate that LUT-Opt reduces subsurface scattering rendering time by approximately 40% and ambient occlusion rendering time by roughly 70%, while incurring only about 2% increase in image quality error, with per-frame inference latency below 0.1\ ms.

We introduce Nemotron 3 Nano Omni, the latest model in the Nemotron multimodal series and the first to natively support audio inputs alongside text, images, and video. Nemotron 3 Nano Omni delivers consistent accuracy improvements over its predecessor, Nemotron Nano V2 VL, across all modalities, enabled by advances in architecture, training data and recipes. In particular, Nemotron 3 delivers leading results in real-world document understanding, long audio-video comprehension, and agentic computer use. Built on the highly efficient Nemotron 3 Nano 30B-A3B backbone, Nemotron 3 Nano Omni further incorporates innovative multimodal token-reduction techniques to deliver substantially lower inference latency and higher throughput than other models of similar size. We are releasing model checkpoints in BF16, FP8, and FP4 formats, along with portions of the training data and codebase to facilitate further research and development.

Large Language Models (LLMs) have achieved strong performance across natural language and multimodal tasks, yet their practical deployment remains constrained by inference latency and kernel launch overhead, particularly in interactive, short-sequence settings. This paper presents a hybrid runtime framework that combines Just-In-Time (JIT) compilation with CUDA Graph execution to reduce launch overhead while preserving runtime flexibility during autoregressive decoding. The framework partitions transformer inference into static components executed via CUDA Graph replay and dynamic components handled through JIT-compiled kernels, enabling asynchronous graph capture and reuse across decoding steps. We evaluate the proposed approach on LLaMA-2 7B using single-GPU, batch-size-one inference across prompt lengths from 10 to 500 tokens. Experimental results show that the hybrid runtime reduces Time-to-First-Token (TTFT) by up to 66.0% and achieves lower P99 latency compared with TensorRT-LLM in this regime. These results indicate that hybrid JIT-CUDA Graph execution can effectively reduce inference latency and variance for short-sequence LLM workloads, making it a practical optimization strategy for latency-sensitive AI applications.

Weeds compete with crops for light, water, and nutrients, reducing yield and crop quality. Efficient weed detection is essential for site-specific weed management (SSWM). Although deep learning models have been deployed on UAV-based edge systems, a systematic understanding of how different model architectures perform under real-world resource constraints is still lacking. To address this gap, this study proposes a deployment-oriented framework for real-time UAV-based weed detection on resource-constrained edge platforms. The framework integrates UAV data acquisition, model development, and on-device inference, with a focus on balancing detection accuracy and computational efficiency. A diverse set of state-of-the-art object detection models is evaluated, including convolution-based YOLO models (v8-v12) and transformer-based RT-DETR models (v1-v2). Experiments on three edge devices (Jetson Orin Nano, Jetson AGX Xavier, and Jetson AGX Orin) demonstrate clear trade-offs between accuracy and inference latency across models and hardware configurations. Results show that high-capacity models achieve up to 86.9% mAP50 but suffer from high latency, limiting real-time deployment. In contrast, lightweight models achieve 66%-71% mAP50 with significantly lower latency, enabling real-time performance. Among all models, RT-DETRv2-R50-M achieves competitive accuracy (79% mAP50) with improved efficiency, while YOLOv10n provides the fastest inference speed. YOLOv11s and RT-DETRv2-R50-M offer the best balance between accuracy and speed, making them strong candidates for real-time UAV deployment.

Large Language Models (LLMs) have demonstrated exceptional performance across diverse domains but are increasingly constrained by high inference latency. Early Exit has emerged as a promising solution to accelerate inference by dynamically bypassing redundant layers. However, in decoder-only architectures, the efficiency of Early Exit is severely bottlenecked by the KV Cache Absence problem, where skipped layers fail to provide the necessary historical states for subsequent tokens. Existing solutions, such as recomputation or masking, either introduce significant latency overhead or incur severe precision loss, failing to bridge the gap between theoretical layer reduction and practical wall-clock speedup. In this paper, we propose River-LLM, a training-free framework that enables seamless token-level Early Exit. River-LLM introduces a lightweight KV-Shared Exit River that allows the backbone's missing KV cache to be naturally generated and preserved during the exit process, eliminating the need for costly recovery operations. Furthermore, we utilize state transition similarity within decoder blocks to predict cumulative KV errors and guide precise exit decisions. Extensive experiments on mathematical reasoning and code generation tasks demonstrate that River-LLM achieves 1.53 to 2.16 times of practical speedup while maintaining high generation quality.

Large reasoning models (LRMs) enhance problem-solving capabilities by generating explicit multi-step chains of thought (CoT) reasoning; however, they incur substantial inference latency and computational overhead. To mitigate this issue, recent works have explored model collaboration paradigms, where small reasoning models (SRMs) generate intermediate reasoning steps to achieve a better accuracy--latency trade-off. Despite recent progress, effectively and efficiently detecting and mitigating SRM failures in collaborative systems remains a key challenge. To address this issue, we analyze SRM inference in both the generated text and hidden-state spaces, and identify three types of failure modes: \textit{overconfidence}, \textit{uncertainty}, and \textit{heavy revalidation}. Building on these insights, we propose \textbf{RankGuide}, a framework that improves the efficiency and effectiveness of SRM--LRM collaboration through tensor-rank-guided routing and steering. Specifically, RankGuide leverages a routing signal that incorporates tensor-rank signals derived from consecutive hidden states to detect when SRMs are likely to fail and selectively invoke LRMs. In addition, we introduce a tensor-rank-filtered steering vector extraction method to modulate the reasoning trajectory of SRMs, thereby improving their generation quality. By improving both routing and steering through tensor-rank signals, RankGuide enables SRM--LRM collaborative systems to achieve more efficient reasoning with fewer steps and improved accuracy. Experiments on multiple reasoning benchmarks demonstrate the efficacy of RankGuide in reducing latency by up to $1.75\times$ compared to LRM, while maintaining competitive accuracy relative to prior methods.

Recent end-to-end spoken dialogue models enable natural interaction. However, as user demands become increasingly complex, models that rely solely on conversational abilities often struggle to cope. Incorporating agentic capabilities is therefore essential: by enabling tool use, these models can extend their knowledge boundaries and better solve real-world tasks. Yet, existing research has largely concentrated on core perception and generation, with comparatively limited exploration of such tool-augmented extensions. To bridge this gap, we present VoxMind, an integrated framework designed to equip end-to-end spoken dialogue models with comprehensive agentic abilities. Leveraging our curated 470-hour AgentChat dataset, we incorporate a "Think-before-Speak" mechanism, enabling the model to internalize structured reasoning as a critical prerequisite for planning and response generation. Furthermore, to mitigate latency bottlenecks caused by large-scale tool integration, we propose a Multi-Agent Dynamic Tool Management architecture. By asynchronously delegating retrieval tasks to an auxiliary agent aligned with the main model's reasoning trajectory, this system effectively decouples inference latency from toolset size. Experimental results confirm that VoxMind achieves significant improvements in agent performance: compared with strong baselines, the task completion rate increases from 34.88% to 74.57%, outperforming Gemini-2.5-Pro on spoken agent tasks while preserving general conversational quality. The source code and associated data are publicly available at https://github.com/MM-Speech/VoxMind.

Autoregressive decoding in Large Language Models (LLMs) generates one token per step, causing high inference latency. Speculative decoding (SD) mitigates this through a guess-and-verify strategy, but existing training-free variants face trade-offs: retrieval-based drafts break when no exact match exists, while logits-based drafts lack structural guidance. We propose $\textbf{RACER}$ ($\textbf{R}$etrieval-$\textbf{A}$ugmented $\textbf{C}$ont$\textbf{e}$xtual $\textbf{R}$apid Speculative Decoding), a lightweight and training-free method that integrates retrieved exact patterns with logit-driven future cues. This unification supplies both reliable anchors and flexible extrapolation, yielding richer speculative drafts. Experiments on Spec-Bench, HumanEval, and MGSM-ZH demonstrate that RACER consistently accelerates inference, achieving more than $2\times$ speedup over autoregressive decoding, and outperforms prior training-free methods, offering a scalable, plug-and-play solution for efficient LLM decoding. Our source code is available at $\href{https://github.com/hkr04/RACER}{https://github.com/hkr04/RACER}$.

Large Reasoning Models (LRMs) achieve strong performance on complex tasks through extended chains of thought but suffer from high inference latency due to autoregressive reasoning. Recent work explores using Small Reasoning Models (SRMs) to accelerate LRM inference. In this paper, we systematically characterize the capability boundaries of SRMs and identify three common types of reasoning risks: (1) path divergence, where SRMs lack the strategic ability to construct an initial plan, causing reasoning to deviate from the most probable path; (2) cognitive overload, where SRMs fail to solve particularly difficult steps; and (3) recovery inability, where SRMs lack robust self-reflection and error correction mechanisms. To address these challenges, we propose TrigReason, a trigger-based collaborative reasoning framework that replaces continuous polling with selective intervention. TrigReason delegates most reasoning to the SRM and activates LRM intervention only when necessary-during initial strategic planning (strategic priming trigger), upon detecting extraordinary overconfidence (cognitive offload trigger), or when reasoning falls into unproductive loops (intervention request trigger). The evaluation results on AIME24, AIME25, and GPQA-D indicate that TrigReason matches the accuracy of full LRMs and SpecReason, while offloading 1.70x - 4.79x more reasoning steps to SRMs. Under edge-cloud conditions, TrigReason reduces latency by 43.9% and API cost by 73.3%. Our code is available at \href{https://github.com/QQQ-yi/TrigReason}{https://github.com/QQQ-yi/TrigReason}

Deploying deep neural networks on edge devices requires balancing accuracy, latency, and resource constraints under realistic execution conditions. To fit models within these constraints, two broad strategies have emerged: static compression techniques such as pruning and quantization, which permanently reduce model size, and dynamic approaches such as early-exit mechanisms, which adapt computational cost at runtime. While both families are widely studied in isolation, they are rarely compared under identical conditions on physical hardware. This paper presents a unified deployment-oriented comparison of static compression and dynamic early-exit mechanisms, evaluated on real edge devices using ONNX based inference pipelines. Our results show that static and dynamic techniques offer fundamentally different trade-offs for edge deployment. While pruning and quantization deliver consistent memory footprint reduction, early-exit mechanisms enable input-adaptive computation savings that static methods cannot match. Their combination proves highly effective, simultaneously reducing inference latency and memory usage with minimal accuracy loss, expanding what is achievable at the edge.

Vision Transformer (ViT)-based sparse multi-view 3D object detectors have achieved remarkable accuracy but still suffer from high inference latency due to heavy token processing. To accelerate these models, token compression has been widely explored. However, our revisit of existing strategies, such as token pruning, merging, and patch size enlargement, reveals that they often discard informative background cues, disrupt contextual consistency, and lose fine-grained semantics, negatively affecting 3D detection. To overcome these limitations, we propose SEPatch3D, a novel framework that dynamically adjusts patch sizes while preserving critical semantic information within coarse patches. Specifically, we design Spatiotemporal-aware Patch Size Selection (SPSS) that assigns small patches to scenes containing nearby objects to preserve fine details and large patches to background-dominated scenes to reduce computation cost. To further mitigate potential detail loss, Informative Patch Selection (IPS) selects the informative patches for feature refinement, and Cross-Granularity Feature Enhancement (CGFE) injects fine-grained details into selected coarse patches, enriching semantic features. Experiments on the nuScenes and Argoverse 2 validation sets show that SEPatch3D achieves up to \textbf{57%} faster inference than the StreamPETR baseline and \textbf{20%} higher efficiency than the state-of-the-art ToC3D-faster, while preserving comparable detection accuracy. Code is available at https://github.com/Mingqj/SEPatch3D.

Routing algorithms are crucial for efficient computer network operations, and in many settings they must be able to react to traffic bursts within milliseconds. Live telemetry data can provide informative signals to routing algorithms, and recent work has trained neural networks to exploit such signals for traffic-aware routing. Yet, aggregating network-wide information is subject to communication delays, and existing neural approaches either assume unrealistic delay-free global states, or restrict routers to purely local telemetry. This leaves their deployability in real-world environments unclear. We cast telemetry-aware routing as a delay-aware closed-loop control problem and introduce a framework that trains and evaluates neural routing algorithms, while explicitly modeling communication and inference delays. On top of this framework, we propose LOGGIA, a scalable graph neural routing algorithm that predicts log-space link weights from attributed topology-and-telemetry graphs. It utilizes a data-driven pre-training stage, followed by on-policy Reinforcement Learning. Across synthetic and real network topologies, and unseen mixed TCP/UDP traffic sequences, LOGGIA consistently outperforms shortest-path baselines, whereas neural baselines fail once realistic delays are enforced. Our experiments further suggest that neural routing algorithms like LOGGIA perform best when deployed fully locally, i.e., observing network states and inferring actions at every router individually, as opposed to centralized decision making.

\emph{Integrated communication and computation} (IC$^2$) has emerged as a new paradigm for enabling efficient edge inference in sixth-generation (6G) networks. However, the design of IC$^2$ technologies is hindered by the lack of a tractable theoretical framework for characterizing \emph{end-to-end} (E2E) inference performance. The metric is highly complicated as it needs to account for both channel distortion and artificial intelligence (AI) model architecture and computational complexity. In this work, we address this challenge by developing a tractable analytical model for E2E inference accuracy and leveraging it to design a \emph{channel-adaptive AI} algorithm that maximizes inference throughput, referred to as the edge processing rate (EPR), under latency and accuracy constraints. Specifically, we consider an edge inference system in which a server deploys a backbone model with early exit, which enables flexible computational complexity, to perform inference on data features transmitted by a mobile device. The proposed accuracy model characterizes high-dimensional feature distributions in the angular domain using a Mixture of von Mises (MvM) distribution. This leads to a desired closed-form expression for inference accuracy as a function of quantization bit-width and model traversal depth, which represents channel distortion and computational complexity, respectively. Building upon this accuracy model, we formulate and solve the EPR maximization problem under joint latency and accuracy constraints, leading to a channel-adaptive AI algorithm that achieves full IC$^2$ integration. The proposed algorithm jointly adapts transmit-side feature compression and receive-side model complexity according to channel conditions to maximize overall efficiency and inference throughput. Experimental results demonstrate its superior performance as compared with fixed-complexity counterparts.

Large-scale Vision-Language-Action (VLA) models offer semantic generalization but suffer from high inference latency because they adopt a low-frequency batch-and-execute paradigm. This frequency mismatch creates an execution blind spot, causing failures in dynamic environments where targets move during the open-loop execution window. We propose TIDAL (Temporally Interleaved Diffusion and Action Loop), a hierarchical framework that decouples semantic reasoning from high-frequency actuation. TIDAL operates as a backbone-agnostic scheduler for diffusion-based VLAs, using a dual-frequency architecture to redistribute the computational budget. Specifically, a low-frequency macro-intent loop caches semantic embeddings, while a high-frequency micro-control loop interleaves single-step flow integration with execution, conditioning on the latest state fused with motion cues. To handle the resulting latency shift, we introduce a temporally misaligned training strategy where the policy learns stalenessaware compensation, conditioning on stale semantic intent alongside real-time proprioception. TIDAL is architectural, making it orthogonal to system-level optimizations. Experiments show an average 2.5x performance gain over open-loop baselines in dynamic interception tasks. Despite a slight decrease in static success rates, our approach yields a 4x increase in feedback frequency and extends the effective horizon of semantic embeddings beyond the native action chunk size. Under nonpaused physics and on a real robot, TIDAL remains robust to unmodeled dynamics, while standard open-loop baselines fail due to latency-induced error accumulation.

Inference of standard convolutional neural networks (CNNs) on FPGAs often incurs high latency and a long initiation interval due to the deep nested loops required to densely convolve every input pixel regardless of its feature value. However, input features can be spatially sparse in some image data, where semantic information may occupy only a small fraction of the pixels and most computation would be wasted on empty regions. In this work, we introduce SparsePixels, a framework that implements sparse convolution on FPGAs by selectively retaining and computing on a small subset of active pixels while ignoring the rest. Because computation always runs over a single pre-specified pixel budget, the inference latency is independent of the input sparsity and is constant at runtime. We show that, for identifying neutrino interactions in naturally sparse LArTPC images with 4k pixels, a standard CNN with a compact size of 4k parameters incurs an inference latency of 48.665 $μ$s on an FPGA, whereas a sparse CNN of the same base architecture, computing on less than 1% of the input pixels, achieves a $\times 73$ speedup to 0.665 $μ$s with resource utilization well within on-chip budgets, trading only a small percent-level performance loss. This work aims to benefit future algorithm development for efficient data readout in modern experiments with latency requirements of microseconds or below.

Scaling inference-time computation has enabled Large Language Models (LLMs) to achieve strong reasoning performance, but their inherently sequential decoding incurs substantial latency, motivating parallelization of the generation process. However, existing parallel reasoning approaches suffer from performance degradation compared to their sequential counterparts, and often rely on specialized inference engines. We introduce ThreadWeaver, a framework for adaptive parallel reasoning that matches the accuracy of comparably sized sequential reasoning models while significantly reducing inference latency via three key innovations: 1) a two-stage parallel trajectory generator that produces high-quality parallel chain-of-thought data for supervised fine-tuning; 2) a trie-based rollout design that enables parallel reasoning on any off-the-shelf autoregressive inference engine; and 3) a parallelization-aware reinforcement learning framework that trains the model to balance reasoning accuracy with effective parallelization. Across six challenging math reasoning benchmarks, ThreadWeaver trained on top of Qwen3-8B achieves performance on par with cutting-edge sequential reasoning models (79.9% on AIME24 and 71.9% on average) while delivering up to 1.53x speedup in token latency, establishing a new Pareto frontier between accuracy and efficiency.

Accurate uncertainty quantification remains a key challenge for standard LLMs, prompting the adoption of Bayesian and ensemble-based methods. However, such methods typically necessitate computationally expensive sampling, involving multiple forward passes to effectively estimate predictive uncertainty. In this paper, we introduce an approach enabling uncertainty estimation in LLMs without incurring the heavy inference latency typically associated with sampling methods. Specifically, we distill uncertainty-aware teachers - originally requiring multiple forward passes - into single-pass students, fine-tuned using LoRA. We compare two distinct distillation strategies: one in which the student employs traditional softmax-based outputs, and another in which the student leverages Dirichlet-distributed outputs to explicitly model epistemic uncertainty via evidential learning. Empirical evaluation on classification tasks demonstrate that such students can achieve comparable predictive and uncertainty quantification performance relative to their teachers, while requiring only a single forward pass.

Browser-based webcam gaze trackers are increasingly used for crowd-scale data collection and in clinical settings where lab eye trackers are impractical, but the reported latency numbers may not represent real world functionality. The common practice of timestamping each gaze sample when it is emitted, rather than when its source frame was captured, makes the measured inference latency read about $0\,$ms no matter how slow the engine really is. We show how to measure it honestly, recovering a per-frame capture clock from the browser's \texttt{re-quest-Video-Frame-Call-back} (rVFC) API (\texttt{captureTime} where the browser exposes it for local camera streams, else \texttt{presentationTime}, in which case every recovered latency is a verifiable lower bound): exact source-frame pairing through a per-frame queue for engines that expose their inference pipeline, and a further lower bound for engines that do not, such as WebGazer. We release an open TypeScript implementation and benchmark harness, demonstrated on two interchangeable engines: WebGazer and a new FaceMesh+KRR pipeline.