~ similar to 2608.15002· 20 results
This paper presents a carbon modeling framework for Processing-in-DRAM (PID) accelerators and evaluates their carbon footprint for various workloads and hardware configurations.
Spotlight reduces the cost of reinforcement learning post-training of Diffusion Transformers by utilizing spot GPUs and reconfiguring sequence parallelism.
This paper demonstrates that Large Language Models (LLMs) can serve as accurate and selective surrogates for costly GPU kernel performance measurements, significantly expanding the search space for op…
This paper presents Isaac-Net, a GPU-batched 5G New Radio module that simulates thousands of environments in lockstep with Isaac Lab physics.
The authors refactor AI and ML algorithms to fit small microcontrollers, achieving efficient inference on low-power devices.
This paper presents NeuronFabric, a software reference architecture for future FPGA and ASIC implementations of transformer training with local Adam updates.
Daifeng Li, Huiqiang Jiang, Chengruidong Zhang, Wei Wu +5 more
The paper introduces D2K-Bench, a diagnostic benchmark to measure how effectively large language model agents translate expert design guidance into efficient GPU kernels.
Haozhe Wang, Lina Chen, Fanqi Wan, Qi Han +4 more
This paper presents Metal Sandbox, a system for efficient on-device inference on macOS that combines restricted file access, native command execution, and command-level scheduling.
The paper introduces Rotary GPU, an exploratory execution approach demonstrating that large Mixture-of-Experts models can be run locally on consumer GPUs with limited VRAM, achieving usable decode thr…
First implementation of a 3D Gaussian renderer on an Intelligence Processing Unit (IPU) with 1,472 independent tiles using only on-chip SRAM.
Tianjiao Yu, Zhilin Ou, Yifan Shen, Xinzhuo Li +4 more
The paper presents CoDA, a framework for co-evolving simulation data and an action policy to improve robot environment construction and policy learning.
Hawkeye is a system that allows perfect, precision-preserving reproduction of GPU-level matrix multiplication operations on a CPU, enabling efficient and trustworthy third-party auditing of machine le…
Paul Koch, Vivek Chavan, André Sers, Adem Karakurt +3 more
This paper discusses current limits and trends in AI vision models for cognitive robotics and proposes a method to bridge the domain gap between simulations and real-world applications.
This paper characterizes the performance of vision-language model inference on the Qualcomm SM8750 using FastVLM-0.5B as a case study, showing significant speedups and energy savings for different pha…
Yangshen Deng, Jiaqi Ruan, Yeqi Huang, Nick Lau +2 more
The paper introduces InfraDev, a benchmark for evaluating AI infrastructure development, and shows that existing benchmarks are inadequate for assessing real-world tasks.
The paper presents a field study on deploying large inference workloads on a non-GPU AI accelerator and identifies eight categories of limitations.