Despite the success of discrete diffusion language models (dLMs) for fast parallel decoding, their non-smooth, high-dimensional space hinders trajectory steering for reasoning and inference acceleration. To overcome this, we present Nemotron-Labs-Sigma, the first family of large-scale (3B/8B) continuous dLMs built on steerable, low-dimensional ODE/SDE latent trajectories. Trained blockwise via likelihood optimization, Nemotron-Labs-Sigma jointly denoises Gaussian-corrupted token embeddings while learning an optimal embedding geometry, and leverages pretrained autoregressive (AR) weights for warm-starting. At inference, we identify classifier-free guidance and score temperature as essential for high-fidelity reasoning and coding. Against state-of-the-art discrete counterparts (masked dLMs and AR baselines), Nemotron-Labs-Sigma outperforms them on standard benchmarks after pretraining (e.g., GSM8K, Minerva, HumanEval, MBPP) and on challenging reasoning tasks after supervised fine-tuning (e.g., MATH-500, AIME), with the largest consistent gains on coding. Beyond accuracy, we exploit two structural properties unique to continuous dLMs. First, smooth trajectories degrade gracefully at low NFEs and enable efficient few-step distillation: distribution matching proves strongest among the methods we study, giving roughly 8x acceleration on math and 4x on coding (8 and 4 tokens per forward pass) with negligible accuracy loss and preserved diversity. In addition, the one-step speculative decoding on the (pre-SFT) distilled 8B model yields a further ~3.5x throughput gain over the AR counterpart (on GB200), outperforming masked dLM baselines. Second, embedding-space steering controls the quality--diversity trade-off, yielding strong pass@$k$ that keeps improving as $k$ grows. Together, these results establish continuous dLMs as a promising paradigm for efficient language generation.
Deep neural networks for image processing are increasingly deployed on resource-constrained edge devices such as drones, medical systems, and smartphones, where limited computational resources and memory bandwidth pose significant challenges for efficient inference. Pruning during training often produces highly sparse convolutional kernels, offering the potential to significantly reduce computation and memory requirements. However, exploiting this sparsity efficiently at runtime remains challenging due to irregular memory access patterns and workload imbalance across processing elements. In this work, we investigate the standard convolution with sparse weights and analyze its performance characteristics through roofline analysis, showing that a 3 x 3 sparse convolution achieves an arithmetic intensity of approximately 10 FLOP/byte compared to roughly 96 FLOP/byte for its dense counterpart. We propose techniques for an efficient sparse convolution at high sparsities. Preliminary results demonstrate that our approach achieves upto 1.65x speedup as compared to dense convolution on NVIDIA A100 and upto 1.52x on NVIDIA Jetson Nano (Tegra X1) , motivating further exploration of direct convolution as a practical path to efficient sparse inference on modern accelerators/edge devices.
Neural surrogates for computational fluid dynamics (CFD) are trained and ranked on datasets that each vary one design family in one flow domain. Such datasets cannot test the least that an aerodynamic foundation model must do: predict CFD-quality incompressible flow in seconds for new shapes in new enclosures. On DrivAerML, larger models and longer training barely lower Transolver's error; doubling the training cases cuts it by about 12% up to 400 cases, exhausting the dataset. We introduce AeroZoo, a benchmark and training corpus of 3,815 time-mean solutions to the unsteady Reynolds-averaged Navier--Stokes (RANS) equations with a transition model, across seven geometry families, eight enclosure configurations and Reynolds numbers from $10^3$ to $10^8$. Each sample poses a well-posed PDE problem in the form a solver receives it: the boundary meshes grouped by condition type, the prescribed values and the fluid properties. The 2,816 samples with a stable time mean form the canonical benchmark: two tasks, engineering metrics and 22 frozen splits, disjoint in geometry, that hold out geometry families, enclosures, boundary-condition layouts and Reynolds ranges. On the random split, boundary-condition tokens cut GeoTransolver's median velocity error from 76.9% to 8.4% of the inlet speed in single-seed runs (uniform inlet flow: 11.0%). ISLA, a transformer for boundary-driven PDEs, reaches 5.3% over four seeds, about 4 times the change from a finer mesh. On the same test samples, a held-out geometry family barely changes ISLA's error relative to uniform flow, and a held-out enclosure or boundary-condition layout raises it 2.4 to 9.2 times. The data and the scoring code will be released.
Federated Learning (FL) with Parameter-Efficient Fine-Tuning (PEFT) enables efficient adaptation of Large Language Models (LLMs), yet its energy and carbon costs across hardware remain poorly understood. We evaluated federated Low-Rank Adaptation (LoRA) fine-tuning on NVIDIA A40, RTX 6000 Ada, and A100 GPUs using a $2^2$ full factorial design with 135M/360M models and 10/100 local steps. CodeCarbon tracked energy and estimated CO$_2$-equivalent emissions. Descriptive allocation of variation assigned 84.8-91.2% of energy and emissions variation to local steps within these ranges. On A40 and RTX 6000 Ada, configurations with fewer steps achieved comparable accuracy with roughly 6-8-fold lower energy consumption and higher accuracy per kWh. These findings support smaller local budgets for energy-efficient federated fine-tuning within the evaluated setting.
Simulating Vein Compressibility for Ultrasound-Guided Femoral Access: Speed, Accuracy, and Phantom Validation
Amit Manicka, Laurent Coulon, Darwin Ang, Michael Loushin, Timothy M. Kowalewski
AI 速读 基于摘要
作者在 NVIDIA Isaac Sim 中测试基于物理的静脉压缩仿真,评估其速度与精度能否支撑机器人超声引导股动脉穿刺训练。
cs.RO
查看原文摘要
In ultrasound-guided femoral access, probe pressure deforms the tissue and changes the vascular anatomy a robot observes. A simulator must reproduce these mechanics fast enough for robotic interaction, yet the computational effort needed for a clinically meaningful signal like vein collapse before artery is unclear. We test the feasibility of a physics-based simulation in NVIDIA's Isaac Sim for Healthcare, with Stable Neo-Hookean tissue solved by Vertex Block Descent on one GPU. A simulation of a low-fidelity cylindrical benchtop phantom with known geometry and mechanical properties, pressed by a robot, predicted its artery within 5 percentage points at every depth. It matched the vein within 3 points through 11 mm of a 25.4 mm press and followed the collapse behavior of the low-fidelity physical phantom. Across element size, time step, and solver iterations, wall clock time per frame scaled linearly with solver work, but accuracy did not improve with more work. The chosen setting reduced solver time per frame by a factor of 976, from 29,700 ms to 30.4 ms, while remaining within 1.03 points of the finest reference setting. The physics solver alone took 30.4 ms per frame, while the full stack, including ultrasound and graphics rendering, took 56.4 ms per frame. Finally, when seeded with realistic literature-based mechanical properties, the digital leg model compressed the vein far more than the artery at every setting, reproducing the expected clinical signal of vein collapse before artery, which is used to distinguish them. In total, these results suggest that the NVIDIA Isaac Sim platform may be suitable for augmenting reinforcement learning training or online simulation for autonomous femoral access with ultrasound.
Spatial-IQ: Deconstructing Spatial Intelligence via Hierarchical Capability Tests
Patrick Rim, Tom Long, Ekta Prashnani, Ruth Rosenholtz, Ben Boudaoud, Peter Xenopoulos, Alex Wong, Joohwan Kim 等 9 位作者
AI 速读 基于摘要
提出分层诊断框架 Spatial-IQ,将堆叠 3D 结构中的物体计数拆解为 9 个感知与认知子任务,用于定位多模态大模型空间推理失败的具体环节。
cs.CV · cs.AI
查看原文摘要
Multimodal large language models (MLLMs) excel at visual interpretation but fail on spatial reasoning tasks that humans solve reliably. Existing benchmarks evaluate these models as black boxes, limiting their ability to identify the underlying causes of lower performance: when a model fails a spatial reasoning task, it remains difficult to ascertain whether the hurdle is perceptual, such as recognizing object boundaries, or cognitive, such as reasoning about occlusion to infer hidden geometry. We introduce Spatial-IQ, a hierarchical diagnostic framework that decomposes object counting in stacked 3D structures into 9 perceptual and cognitive sub-tasks organized by the developmental stages of human spatial cognition, with mental rotation as an additional target probe. Using NVIDIA Isaac Sim, we procedurally generated a diverse dataset of roughly 80,000 stacked 3D structures with per-task ground truth. We evaluate models across three output formats (free-response text, multiple-choice images, and image editing) alongside a human baseline. The Spatial-IQ framework shows that top-performing models often succeed at the target task (object counting) without succeeding on the lower-level sub-tasks intended to support it, and that models differ in how much of these hierarchical chains they preserve, often revealing shortcut behavior that raw target-task accuracy alone would obscure. Finally, we demonstrate that training models with chain-of-thought (CoT) supervision over our hierarchical sub-tasks, combined with reinforcement learning with verifiable rewards, significantly improves both spatial consistency across sub-tasks and target-task accuracy, supporting the value of the proposed decomposition as both a diagnostic tool and a training signal.
The key-value (KV) cache dominates memory capacity and bandwidth in long-context LLM serving. While 4-bit microscaling formats (NVFP4, MXFP4) shrink the cache and run attention on FP4 Tensor Cores, a fundamental gap remains between algorithmic quantization and system-aware acceleration: existing FP4 KV systems use vanilla quantization that causes severe accuracy drops, whereas accuracy-oriented 4-bit algorithms rely on vector codebooks whose lookup overhead degrades decoding throughput. We present BitDecoding-FP4, an algorithm--system co-design for accurate and efficient native FP4 KV decoding. We mitigate three error sources: Dual-Residual Query Quantization splits queries into two FP4 tensors, cutting query logit error from 60% to 1%; Underflow-Aware Probability Scaling prevents E4M3 scales from rounding long-context probabilities to zero; and Outlier-Aware Key Quantization combines static smoothing, Hadamard rotation, and scale search to suppress key outliers under token-wise scaling. At the system level, we remove the softmax bottleneck via transposed softmax with fused probability quantization and stacked query MMA on paged KV caches. Across dense and MoE models with GQA and MLA, BitDecoding-FP4 stays within 0.5 points of BF16 KV accuracy with NVFP4 (1 point with MXFP4), and tracks BF16 KV in RL training rollouts. In end-to-end serving on NVIDIA B300 with Qwen3-4B and Qwen3-30B-A3B, it delivers 2.4-3.0$\times$ higher peak serving throughput than BF16 KV. The implementation is available at https://github.com/OpenBitSys/BitDecoding-FP4.
Beyond Benchmarks: Continuous Edge Inference for Traffic Sign Recognition on Embedded Vehicular Hardware
Aditya Mishra, Anushka Tailor, Haroon Lone
AI 速读 基于摘要
作者提出面向NVIDIA Jetson Orin Nano的连续边缘推理系统Edge-TSR,并在流式视频评测中揭示图像基准与真实部署之间的性能落差。
cs.CV · cs.RO · eess.SY
查看原文摘要
Traffic sign recognition (TSR) systems deployed on embedded vehicular hardware face deployment effects that image-based benchmark evaluation does not capture, including frame-to-frame prediction instability in streaming video and load-dependent throughput under sustained operation. We present Edge-TSR, a continuous edge inference system for TSR on the NVIDIA Jetson Orin Nano that combines sparse detection, multi-object tracking, region-level fine-grained classification, and a lightweight track-aware temporal stabilization mechanism adding only 0.07~ms per sampled frame. Evaluating on the source videos of the INTSD benchmark, we find that the state-of-the-art TSR models lose 20-30 percentage points of mAP@50 when moving from image-based to streaming evaluation. Edge-TSR achieves the best streaming classification performance among all evaluated baselines, and temporal stabilization improves accuracy by 10.58 percentage points over per-frame inference. We characterize latency, throughput, and thermal behavior across sampling intervals and four driving conditions, showing that throughput is load-proportional (20 ms per classified region) and that every-frame inference is limited by throughput rather than thermal throttling. A 55-minute, 26 km vehicular deployment demonstrates sustained online operation at 16.18 FPS within safe thermal limits without cloud offload. We also release the annotated 50,732-frame streaming evaluation dataset and full implementation.
ML code should run where researchers choose, not where it was first written. Today, machine learning code mainly lives in two ecosystems, PyTorch on NVIDIA GPUs and JAX on Google TPUs. Migrating repositories between them means building and testing across frameworks and hardware. This migration is (1) a practical need, since it unlocks different hardware for researchers, and (2) a natural benchmark: it involves different frameworks and hardware, covers whole repositories, and is long-horizon and challenging. We introduce XStack, a benchmark that evaluates agents on migrating ML repositories between these two ecosystems. Each agent receives a PyTorch repository and a clean TPU machine and must deliver a JAX repository. Its output passes if it meets our grading criteria, based on the source repository's published result, authenticated and calibrated by us. Migration by agents remains unreliable: Claude Opus 5 passes 75%, GPT-5.6-sol 45%, and every other agent at most 40%. XStack is a step toward freeing ML code from its ecosystem. We release XStack to make progress toward reliable migration across frameworks and hardware measurable.
Road anomaly detection systems on autonomous vehicles must function reliably under degraded visual conditions including darkness, rain, and Harmattan dust haze. This paper benchmarks four object detection architectures, YOLOv8n, YOLOv12n, RT-DETR, and Faster R-CNN, under physically grounded lighting and weather degradation, and shows that uniform application of the SCI low-light enhancement network is actively harmful at dusk-level lighting (-16.1% to -30.7% mAP@50). We propose a lightweight, training-free luminance-adaptive router that selects between no enhancement, moderate SCI, and denoising plus strong SCI based on per-frame Rec. 601 luminance, recovering most of the dusk penalty while preserving low-light gains, and we validate the full pipeline on NVIDIA Jetson Nano edge hardware.
Datacenter GPUs increasingly favour low-precision floating-point formats, offering higher throughput than FP64. Double-word (DW) arithmetic represents numbers as unevaluated sums of two floating-point values, providing higher accuracy. We present the first error analysis of complex DW multiplication and division. Both operations reduce to sums of two DW products, combined using one of three DW addition algorithms: accurate, sloppy, or Madd. Normalising DW products ensures nonoverlapping components, but omitting this step reduces the number of arithmetic operations. We bound the resulting overlap and prove that Madd retains a relative error bound of $3u^2+O(u^3)$ for overlapping inputs, where $u$ is the unit roundoff. For all variants, we establish relative error bounds of the form $(ακ+β)u^2+O(u^3)$, where $κ$ is the condition number of the sum of products. Our analysis shows that the less expensive sloppy addition suffices. We implement these variants in CompleXDW, an open-source C++ library for CPUs and GPUs. In the high-energy physics code MadGraph7, FP32-based DW arithmetic achieves up to 2.9x and 2.1x higher throughput than FP64 on NVIDIA L40S and AMD Radeon Pro W7900 GPUs, respectively, while losing approximately 1.4 correct significant digits.
Adaptive rounding methods such as GPTQ quantize the weights of a large language model (LLM) one at a time, propagating each rounding error to the weights not yet rounded. YAQA lowers the KL divergence from the original model with two-sided adaptive rounding under a Kronecker-factored Hessian, but its error feedback takes quartic work in the size of a square layer. We present GPTQ-2D, which provably computes the same rounding with cubic work, while still rounding the entries of each anti-diagonal in parallel. Its GPU implementation uses two levels of blocking to improve hardware utilization. We also extend the algorithm to Hessians with any number of Kronecker factors. In our experiments, GPTQ-2D is as accurate as YAQA and cuts the time to quantize Llama-3.1-70B-Instruct on a single NVIDIA RTX A6000 GPU from 8.6 days to 1.6 hours. Source code is available at https://github.com/IST-DASLab/GPTQ-2D.
We present LucidLyra, a framework that makes generatively reconstructed 3D scenes directly queryable for semantics and instances in a feed-forward manner from video latents. While generative 3D reconstruction can synthesize virtual environments from a single image, downstream applications in robotics and physical AI require not only visual realism but also access to scene semantics and object identities. Existing 3D understanding approaches, however, require captured multi-view scenes, have limited generalization capabilities, or rely on expensive per-scene processing and optimization. To bridge this gap, our method distills complementary knowledge from a camera-controlled video diffusion model and frozen 2D vision foundation models into a unified model to jointly infer 3D Gaussians encoding geometry, appearance, open-vocabulary semantics, and class-agnostic instance cues. The resulting scenes support novel-view rendering, open-vocabulary segmentation, and object grouping without per-scene optimization, while training requires neither manual 3D annotations nor captured multi-view data. Our method outperforms baselines combining state-of-the-art 3DGS segmentation approaches with generative 3D scene reconstruction on curated synthetic and real-image-conditioned benchmarks, while preserving reconstruction quality. Furthermore, our latent formulation naturally extends to monocular video, reconstructing structured dynamic 3D scenes across time. By turning renderable but unstructured outputs into queryable 3D worlds, LucidLyra enables downstream applications including 3D scene editing, 3D asset mining, and robotic simulation.
Reconstructing dynamic 3D scenes from monocular videos remains challenging because limited observations provide insufficient motion constraints. Conventional 4D Gaussian representations model spatiotemporal appearance efficiently, but the restricted temporal evolution of each primitive makes complex nonlinear dynamics difficult to represent compactly. We propose a motion-expressive 4D Gaussian representation with time-varying means and covariances. By analytically differentiating Gaussian trajectories, we obtain Gaussian velocities and introduce Flow Splatting to render optical flow while accounting for camera motion, enabling dense motion supervision. This formulation jointly models scene appearance and motion within a unified set of 4D Gaussian primitives. We further introduce an initialization strategy based on the sampling theorem and a velocity consistency regularizer to stabilize optimization. Experiments on DAVIS and NVIDIA Dynamic Scenes demonstrate a favorable trade-off between reconstruction quality and efficiency under challenging monocular settings.
Power capping is a common GPU energy lever in LLM serving, and it appears to work: the cap is set, the GPU reports drawing less than it, and the energy budget is met. We show the cap does little for the phase that dominates LLM serving: autoregressive decode. Across four attention paradigms (GQA, MLA, Gated DeltaNet, and Mamba2) on an NVIDIA H200 under vLLM eager execution, decode kernels wait on HBM and the GPU idles between kernel launches, so decode draws less than half of the GPU's 700W rating and throughput stays within a few percent under every cap we set. The cap engages only when a large batch runs a long generation. Locking the SM clock instead lowers power without slowing memory-bound decode and recovers up to 32% of decode energy at minimal throughput loss. We identify three architecture-dependent frequency-scaling classes and find that, once prefill is measured without prefix caching, the recurrent architectures pay no prefill penalty: their linear-time mixing undercuts quadratic attention on long prompts, so their efficient decode adds to the prefill saving and lowers total request energy from the first output token at large batch.