We open-sourced Paddock, our Rust/C++ inference engine with its own CUDA kernels (MIT/Apache-2.0)

NVIDIA模型发布模型访问端侧推理

inference

时间与来源

收录: 09/04 21:00
来源类型: 未分类

讨论趋势

→ 平稳

最近 24 小时与此前 24 小时的快照均值对比 · 7 天曲线

百分比基于采集到的讨论信号，不代表新增评论数或独立参与人数。曲线仅用于同一话题在不同时段的比较。

AI 摘要

Paddock, a Rust/C++ inference engine with custom CUDA kernels, has been open-sourced under MIT/Apache-2.0 licenses. Developed internally, it now operates publicly, supporting one model per GPU on CUDA for Windows and Linux. It's validated on Blackwell (5090, RTX PRO 4500/5000/6000, B200) and Ampere (A6000, 30-series), with Ada kernels available via PADDOCK_UNVALIDATED_ARCH=1, and Hopper/A100 kernels in the tree.