This paper investigates the recognition of the Russian fingerspelling alphabet, also known as the Russian Sign Language (RSL) dactyl. Dactyl is a component of sign languages where distinct hand movements represent individual letters of a written language. This method is used to spell words without specific signs, such as proper nouns or technical terms. The alphabet learning simulator is an essential isolated dactyl recognition application. There is a notable issue of data shortage in isolated dactyl recognition: existing Russian dactyl datasets lack subject heterogeneity, contain insufficient samples, or cover only static signs. We provide Bukva, the first full-fledged open-source video dataset for RSL dactyl recognition. It contains 3,757 videos with more than 101 samples for each RSL alphabet sign, including dynamic ones. We utilized crowdsourcing platforms to increase the subject's heterogeneity, resulting in the participation of 155 deaf and hard-of-hearing experts in the dataset creation. We use a TSM (Temporal Shift Module) block to handle static and dynamic signs effectively, achieving 83.6% top-1 accuracy with a real-time inference with CPU only. The dataset, demo code, and pre-trained models are publicly available.
We present \texttt{MathOptAI.jl}, an open-source Julia library for embedding trained machine learning predictors into a JuMP model. \texttt{MathOptAI.jl} can embed a wide variety of neural networks, decision trees, and Gaussian Processes into a larger mathematical optimization model. In addition to interfacing a range of Julia-based machine learning libraries such as \texttt{Lux.jl} and \texttt{Flux.jl}, \texttt{MathOptAI.jl} uses Julia's Python interface to provide support for PyTorch models. When the PyTorch support is combined with \texttt{MathOptAI.jl}'s gray-box formulation, the function, Jacobian, and Hessian evaluations associated with the PyTorch model are offloaded to the GPU in Python, while the rest of the nonlinear oracles are evaluated on the CPU in Julia. \MathOptAI is available at https://github.com/lanl-ansi/MathOptAI.jl under a BSD-3 license.
The recent shift in Generative AI (GenAI) applications from cloud-only environments to end-user devices introduces new challenges in resource management, system efficiency, and user experience. This paper presents ConsumerBench, a comprehensive benchmarking framework designed to evaluate the system efficiency and response time of GenAI models running on end-user devices. Unlike existing benchmarks that assume exclusive model access on dedicated GPUs, ConsumerBench simulates realistic multi-application scenarios executing concurrently on constrained hardware. Furthermore, ConsumerBench supports customizable workflows that simulate complex tasks requiring coordination among multiple applications. ConsumerBench captures both application-level metrics, including latency and Service Level Objective (SLO) attainment, and system-level metrics like CPU/GPU utilization and memory bandwidth. Through extensive experiments, ConsumerBench reveals inefficiencies in resource sharing, unfair scheduling under greedy allocation, and performance pitfalls of static model server configurations. The paper also provides practical insights for model developers and system designers, highlighting the benefits of custom kernels tailored to consumer-grade GPU architectures and the value of implementing SLO-aware scheduling strategies.
We introduce a new dataset and algorithm for fast and efficient coastal distance calculations from Anywhere on Earth (AoE). Existing global coastal datasets are only available at coarse resolution (e.g. 1-4 km) which limits their utility. Publicly available satellite imagery combined with computer vision enable much higher precision. We provide a global coastline dataset at 10 meter resolution, a 100+ fold improvement in precision over existing data. To handle the computational challenge of querying at such an increased scale, we introduce a new library: Layered Iterative Geospatial Hierarchical Terrain-Oriented Unified Search Engine (Lighthouse). Lighthouse is both exceptionally fast and resource-efficient, requiring only 1 CPU and 2 GB of RAM to achieve millisecond online inference, making it well suited for real-time applications in resource-constrained environments.
Location-based services often require users to share sensitive locational data, raising privacy concerns due to potential misuse or exploitation by untrusted servers. In response, we present VeLoPIR, a versatile location-based private information retrieval (PIR) system designed to preserve user privacy while enabling efficient and scalable query processing. VeLoPIR introduces three operational modes-interval validation, coordinate validation, and identifier matching-that support a broad range of real-world applications, including information and emergency alerts. To enhance performance, VeLoPIR incorporates multi-level algorithmic optimizations with parallel structures, achieving significant scalability across both CPU and GPU platforms. We also provide formal security and privacy proofs, confirming the system's robustness under standard cryptographic assumptions. Extensive experiments on real-world datasets demonstrate that VeLoPIR achieves up to 11.55 times speed-up over a prior baseline. The implementation of VeLoPIR is publicly available at https://github.com/PrivStatBool/VeLoPIR.
The widespread adoption of outsourced neural network inference presents significant privacy challenges, as sensitive user data is processed on untrusted remote servers. Secure inference offers a privacy-preserving solution, but existing frameworks suffer from high computational overhead and communication costs, rendering them impractical for real-world deployment. We introduce SecONNds, a non-intrusive secure inference framework optimized for large ImageNet-scale Convolutional Neural Networks. SecONNds integrates a novel fully Boolean Goldreich-Micali-Wigderson (GMW) protocol for secure comparison -- addressing Yao's millionaires' problem -- using preprocessed Beaver's bit triples generated from Silent Random Oblivious Transfer. Our novel protocol achieves an online speedup of 17$\times$ in nonlinear operations compared to state-of-the-art solutions while reducing communication overhead. To further enhance performance, SecONNds employs Number Theoretic Transform (NTT) preprocessing and leverages GPU acceleration for homomorphic encryption operations, resulting in speedups of 1.6$\times$ on CPU and 2.2$\times$ on GPU for linear operations. We also present SecONNds-P, a bit-exact variant that ensures verifiable full-precision results in secure computation, matching the results of plaintext computations. Evaluated on a 37-bit quantized SqueezeNet model, SecONNds achieves an end-to-end inference time of 2.8 s on GPU and 3.6 s on CPU, with a total communication of just 420 MiB. SecONNds' efficiency and reduced computational load make it well-suited for deploying privacy-sensitive applications in resource-constrained environments. SecONNds is open source and can be accessed from: https://github.com/shashankballa/SecONNds.
Multi-head self-attention (MHSA) has become a core component in modern computer vision models. However, its quadratic complexity with respect to input length poses a significant computational bottleneck in real-time and resource constrained environments. We propose PnP-Nystra, a Nystr\"om based linear approximation of self-attention, developed as a plug-and-play (PnP) module that can be integrated into the pre-trained image and video restoration models without retraining. As a drop-in replacement for MHSA, PnP-Nystra enables efficient acceleration in various window-based transformer architectures, including SwinIR, Uformer, and RVRT. Our experiments across diverse image and video restoration tasks, including denoising, deblurring, and super-resolution, demonstrate that PnP-Nystra achieves a 2-4x speed-up on an NVIDIA RTX 4090 GPU and a 2-5x speed-up on CPU inference. Despite these significant gains, the method incurs a maximum PSNR drop of only 1.5 dB across all evaluated tasks. To the best of our knowledge, we are the first to demonstrate a linear attention functioning as a training-free substitute for MHSA in restoration models.
The computational sparsity of Mixture-of-Experts (MoE) models enables sub-linear growth in compute cost as model size increases, thus offering a scalable path to training massive neural networks. However, existing implementations suffer from \emph{low GPU utilization}, \emph{significant latency overhead}, and a fundamental \emph{inability to leverage task locality}, primarily due to CPU-managed scheduling, host-initiated communication, and frequent kernel launches. To overcome these limitations, we develop FlashDMoE, a fully GPU-resident MoE operator that fuses expert computation and inter-GPU communication into a \emph{single persistent GPU kernel}. FlashDMoE enables fine-grained pipelining of dispatch, compute, and combine phases, eliminating launch overheads and reducing idle gaps. Unlike existing work, FlashDMoE obviates bulk-synchronous collectives for one-sided, device-initiated, inter-GPU (R)DMA transfers, thus unlocking \emph{payload efficiency}, where we eliminate bloated or redundant network payloads in sparsely activated layers. When evaluated on a single 8-H100 GPU node with MoE models having up to 128 experts and 16K token sequences, FlashDMoE achieves up to \textbf{9}$\times$ higher GPU utilization, \textbf{6}$\times$ lower latency, \textbf{5.7}$\times$ higher throughput, and \textbf{4}$\times$ better overlap efficiency compared to state-of-the-art baselines, despite using FP32 while baselines use FP16. FlashDMoE demonstrates that principled GPU kernel-hardware co-design is key to unlocking the performance ceiling of large-scale distributed ML workloads.
The Key-Value (KV) cache in generative large language models (LLMs) introduces substantial memory overhead. Existing works mitigate this burden by offloading or compressing the KV cache. However, loading the entire cache incurs significant latency due to PCIe bandwidth bottlenecks in CPU-GPU communication, while aggressive compression causes notable performance degradation. We identify that certain layers in the LLM need to maintain global information and are unsuitable for selective loading. In contrast, other layers primarily focus on a few tokens with dominant activations that potentially incur substantial quantization error. This observation leads to a key insight that loading dominant tokens and quantizing all tokens can complement each other. Building on this insight, we propose a hybrid compression method, TailorKV, which seamlessly integrates quantization and offloading. TailorKV develops an inference framework along with a hardware-friendly implementation that leverages these complementary characteristics. Extensive long-context evaluations exhibit that TailorKV achieves nearly lossless performance under aggressive compression settings, outperforming the state-of-the-art. Particularly, the Llama-3.1-8B with 128k context can be served within a single RTX 3090 GPU, reaching 82 ms per token during decoding.
Robust long-term visual localization in complex industrial environments is critical for mobile robotic systems. Existing approaches face limitations: handcrafted features are illumination-sensitive, learned features are computationally intensive, and semantic- or marker-based methods are environmentally constrained. Handcrafted and learned features share similar representations but differ functionally. Handcrafted features are optimized for continuous tracking, while learned features excel in wide-baseline matching. Their complementarity calls for integration rather than replacement. Building on this, we propose a hierarchical localization framework. It leverages real-time handcrafted feature extraction for relative pose estimation. In parallel, it employs selective learned keypoint detection on optimized keyframes for absolute positioning. This design enables CPU-efficient, long-term visual localization. Experiments systematically progress through three validation phases: Initially establishing feature complementarity through comparative analysis, followed by computational latency profiling across algorithm stages on CPU platforms. Final evaluation under photometric variations (including seasonal transitions and diurnal cycles) demonstrates 47% average error reduction with significantly improved localization consistency. The code implementation is publicly available at https://github.com/linyicheng1/ORB_SLAM3_localization.
Long-video understanding has emerged as a crucial capability in real-world applications such as video surveillance, meeting summarization, educational lecture analysis, and sports broadcasting. However, it remains computationally prohibitive for VideoLLMs, primarily due to two bottlenecks: 1) sequential video decoding, the process of converting the raw bit stream to RGB frames can take up to a minute for hour-long video inputs, and 2) costly prefilling of up to several million tokens for LLM inference, resulting in high latency and memory use. To address these challenges, we propose QuickVideo, a system-algorithm co-design that substantially accelerates long-video understanding to support real-time downstream applications. It comprises three key innovations: QuickDecoder, a parallelized CPU-based video decoder that achieves 2-3 times speedup by splitting videos into keyframe-aligned intervals processed concurrently; QuickPrefill, a memory-efficient prefilling method using KV-cache pruning to support more frames with less GPU memory; and an overlapping scheme that overlaps CPU video decoding with GPU inference. Together, these components infernece time reduce by a minute on long video inputs, enabling scalable, high-quality video understanding even on limited hardware. Experiments show that QuickVideo generalizes across durations and sampling rates, making long video processing feasible in practice.
Mixture-of-Experts (MoE) enables efficient scaling of large language models (LLMs) with sparsely activated experts during inference. To effectively deploy large MoE models on memory-constrained devices, many systems introduce *expert offloading* that caches a subset of experts in fast memory, leaving others on slow memory to run on CPU or load on demand. While some research has exploited the locality of expert activations, where consecutive tokens activate similar experts, the degree of this **local routing consistency** varies across models and remains understudied. In this paper, we propose two metrics to measure local routing consistency of MoE models: (1) **Segment Routing Best Performance (SRP)**, which evaluates how well a fixed group of experts can cover the needs of a segment of tokens, and (2) **Segment Cache Best Hit Rate (SCH)**, which measures the optimal segment-level cache hit rate under a given cache size limit. We analyzed 20 MoE LLMs with diverse sizes and architectures and found that models that apply MoE on every layer and do not use shared experts exhibit the highest local routing consistency. We further showed that domain-specialized experts contribute more to routing consistency than vocabulary-specialized ones, and that most models can balance between cache effectiveness and efficiency with cache sizes approximately 2x the active experts. These findings pave the way for memory-efficient MoE design and deployment without compromising inference speed. We publish the code for replicating experiments at https://github.com/ljcleo/moe-lrc .
Efficient task scheduling is paramount in the Linux kernel, where the Completely Fair Scheduler (CFS) meticulously manages CPU resources to balance high utilization with interactive responsiveness. This research pioneers the use of deep learning techniques to predict the sequence of tasks selected by CFS, aiming to evaluate the feasibility of a more generalized and potentially more adaptive task scheduler for diverse workloads. Our core contributions are twofold: first, the systematic generation and curation of a novel scheduling dataset from a running Linux kernel, capturing real-world CFS behavior; and second, the development, training, and evaluation of a Long Short-Term Memory (LSTM) network designed to accurately forecast the next task to be scheduled. This paper further discusses the practical pathways and implications of integrating such a predictive model into the kernel's scheduling framework. The findings and methodologies presented herein open avenues for data-driven advancements in kernel scheduling, with the full source code provided for reproducibility and further exploration.
Efficient LLM inference on resource-constrained devices presents significant challenges in compute and memory utilization. Due to limited GPU memory, existing systems offload model weights to CPU memory, incurring substantial I/O overhead between the CPU and GPU. This leads to two major inefficiencies: (1) GPU cores are underutilized, often remaining idle while waiting for data to be loaded; and (2) GPU memory has low impact on performance, as reducing its capacity has minimal effect on overall throughput.In this paper, we propose SpecOffload, a high-throughput inference engine that embeds speculative decoding into offloading. Our key idea is to unlock latent GPU resources for storing and executing a draft model used for speculative decoding, thus accelerating inference at near-zero additional cost. To support this, we carefully orchestrate the interleaved execution of target and draft models in speculative decoding within the offloading pipeline, and propose a planner to manage tensor placement and select optimal parameters. Compared to the best baseline, SpecOffload improves GPU core utilization by 4.49x and boosts inference throughput by 2.54x. Our code is available at https://github.com/MobiSense/SpecOffload-public .
Modal methods for simulating vibrations of strings, membranes, and plates are widely used in acoustics and physically informed audio synthesis. However, traditional implementations, particularly for non-linear models like the von K\'arm\'an plate, are computationally demanding and lack differentiability, limiting inverse modelling and real-time applications. We introduce a fast, differentiable, GPU-accelerated modal framework built with the JAX library, providing efficient simulations and enabling gradient-based inverse modelling. Benchmarks show that our approach significantly outperforms CPU and GPU-based implementations, particularly for simulations with many modes. Inverse modelling experiments demonstrate that our approach can recover physical parameters, including tension, stiffness, and geometry, from both synthetic and experimental data. Although fitting physical parameters is more sensitive to initialisation compared to other methods, it provides greater interpretability and more compact parameterisation. The code is released as open source to support future research and applications in differentiable physical modelling and sound synthesis.
In this paper, we present a 3D reconstruction and rendering framework termed Mesh-Learner that is natively compatible with traditional rasterization pipelines. It integrates mesh and spherical harmonic (SH) texture (i.e., texture filled with SH coefficients) into the learning process to learn each mesh s view-dependent radiance end-to-end. Images are rendered by interpolating surrounding SH Texels at each pixel s sampling point using a novel interpolation method. Conversely, gradients from each pixel are back-propagated to the related SH Texels in SH textures. Mesh-Learner exploits graphic features of rasterization pipeline (texture sampling, deferred rendering) to render, which makes Mesh-Learner naturally compatible with tools (e.g., Blender) and tasks (e.g., 3D reconstruction, scene rendering, reinforcement learning for robotics) that are based on rasterization pipelines. Our system can train vast, unlimited scenes because we transfer only the SH textures within the frustum to the GPU for training. At other times, the SH textures are stored in CPU RAM, which results in moderate GPU memory usage. The rendering results on interpolation and extrapolation sequences in the Replica and FAST-LIVO2 datasets achieve state-of-the-art performance compared to existing state-of-the-art methods (e.g., 3D Gaussian Splatting and M2-Mapping). To benefit the society, the code will be available at https://github.com/hku-mars/Mesh-Learner.
High-throughput neural network inference requires coordinating many optimization decisions, including parallel tiling, microkernel selection, and data layout. The product of these decisions forms a search space of programs which is typically intractably large. Existing approaches (e.g., auto-schedulers) often address this problem by sampling this space heuristically. In contrast, we introduce a dynamic-programming-based approach to explore more of the search space by iteratively decomposing large program specifications into smaller specifications reachable from a set of rewrites, then composing a final program from each rewrite that minimizes an affine cost model. To reduce memory requirements, we employ a novel memoization table representation, which indexes specifications by coordinates in $Z_{\geq 0}$ and compresses identical, adjacent solutions. This approach can visit a much larger set of programs than prior work. To evaluate the approach, we developed Morello, a compiler which lowers specifications roughly equivalent to a few-node XLA computation graph to x86. Notably, we found that an affine cost model is sufficient to surface high-throughput programs. For example, Morello synthesized a collection of matrix multiplication benchmarks targeting a Zen 1 CPU, including a 1x2048x16384, bfloat16-to-float32 vector-matrix multiply, which was integrated into Google's gemma.cpp.
Python is the de-facto language for software development in artificial intelligence (AI). Commonly used libraries, such as PyTorch and TensorFlow, rely on parallelization built into their BLAS backends to achieve speedup on CPUs. However, only applying parallelization in a low-level backend can lead to performance and scaling degradation. In this work, we present a novel way of binding task-based C++ code built on the asynchronous runtime model HPX to a high-level Python API using pybind11. We develop a parallel Gaussian process (GP) li- brary as an application. The resulting Python library GPRat combines the ease of use of commonly available GP libraries with the performance and scalability of asynchronous runtime systems. We evaluate the per- formance on a mass-spring-damper system, a standard benchmark from control theory, for varying numbers of regressors (features). The results show almost no binding overhead when binding the asynchronous HPX code using pybind11. Compared to GPyTorch and GPflow, GPRat shows superior scaling on up to 64 cores on an AMD EPYC 7742 CPU for train- ing. Furthermore, our library achieves a prediction speedup of 7.63 over GPyTorch and 25.25 over GPflow. If we increase the number of features from eight to 128, we observe speedups of 29.62 and 21.19, respectively. These results showcase the potential of using asynchronous tasks within Python-based AI applications.
Large Language Models (LLMs) have grown rapidly in size, creating significant challenges for efficient deployment on resource-constrained hardware. In this paper, we introduce Dynamic-Length Float (DFloat11), a lossless compression framework that reduces LLM size by 30% while preserving outputs that are bit-for-bit identical to the original model. DFloat11 is motivated by the low entropy in the BFloat16 weight representation of LLMs, which reveals significant inefficiency in existing storage format. By applying entropy coding, DFloat11 assigns dynamic-length encodings to weights based on frequency, achieving near information-optimal compression without any loss of precision. To facilitate efficient inference with dynamic-length encodings, we develop a custom GPU kernel for fast online decompression. Our design incorporates the following: (i) decomposition of memory-intensive lookup tables (LUTs) into compact LUTs that fit in GPU SRAM, (ii) a two-phase kernel for coordinating thread read/write positions using lightweight auxiliary variables, and (iii) transformer-block-level decompression to minimize latency. Experiments on recent models, including Llama-3.1, Qwen-2.5, and Gemma-3, validates our hypothesis that DFloat11 achieves around 30% model size reduction while preserving bit-for-bit exact outputs. Compared to a potential alternative of offloading parts of an uncompressed model to the CPU to meet memory constraints, DFloat11 achieves 1.9-38.8x higher throughput in token generation. With a fixed GPU memory budget, DFloat11 enables 5.3-13.17x longer context lengths than uncompressed models. Notably, our method enables lossless inference of Llama-3.1-405B, an 810GB model, on a single node equipped with 8x80GB GPUs. Our code and models are available at https://github.com/LeanModels/DFloat11.
Emergency of DeepSeek R1 and QwQ 32B have broken through performance barriers for running frontier large language models (LLMs) on home devices. While consumer hardware is getting stronger and model quantization is improving, existing end-side solutions still demand GPU clusters, large RAM/VRAM, and high bandwidth, far beyond what a common home cluster can handle. This paper introduces prima.cpp, a distributed inference system that runs 70B-scale models on everyday home devices using a mix of CPU/GPU, low RAM/VRAM, Wi-Fi, and cross-platform support. It uses mmap to manage model weights and introduces piped-ring parallelism with prefetching to hide disk loading. By modeling heterogeneity in computation, communication, disk, memory (and its management behavior), and OS, it optimally assigns model layers to each device's CPU and GPU, further reducing token latency. An elegant algorithm named Halda is proposed to solve this NP-hard assignment problem. We evaluate prima.cpp on a common four-node home cluster. It outperforms llama.cpp, exo, and dllama on 30B+ models while keeping memory pressure below 6%. This brings frontier 30B-70B models, such as Llama 3, DeepSeek R1, Qwen 2.5, and QwQ to home assistants, making advanced AI truly accessible to individuals. The code is open source and available at https://github.com/Lizonghang/prima.cpp.