NVIDIA 已开源其 cuDNN 库的部分内容，这是在闭源 12 年后的一项重大举措。此次发布包括 20 多个专家混合 (MoE) 内核和 NSA 稀疏注意力内核。这些内核的代码库大部分是用 Python CuTe-DSL 编写的，现已提供公开文档。 AI
影响 cuDNN 内核的开源可能会加速人工智能基础设施和模型优化的研究与开发。
排序理由 一家主要科技公司开源了重要的软件库组件。
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
NVIDIA 在 12 年后开源 cuDNN 内核，包括 MoE 和稀疏注意力
报道来源 [3]
-
@manicely6005 The public documentation can be found here too (3/3)
@manicely6005 The public documentation can be found here too (3/3) https://t.co/jUnoOJ7kj7 https://t.co/M74p6UyTKc
-
@manicely6005 The codebase for these kernels, a lot of these OSS cuDNN kernels are written in Python CuTe-DSL (2/3)
@manicely6005 The codebase for these kernels, a lot of these OSS cuDNN kernels are written in Python CuTe-DSL (2/3) https://t.co/7awOYa0dbd https://t.co/euu5qaKveW
-
For the past 12 years, cuDNN has been completely closed sourced (besides the .h files), until this week! OVER 20 MoE kernels & NSA sparse attention kernels
For the past 12 years, cuDNN has been completely closed sourced (besides the .h files), until this week! OVER 20 MoE kernels & NSA sparse attention kernels from cuDNN has been open sourced! Great work to @manicely6005 & the rest of the team on seeing that parts of NVIDIA …