NVIDIA 用 CUDA Tile 高性能前向内核加速 Mamba-2

Accelerating Mamba-2 with High-Performance Forward Pass Kernels in NVIDIA CUDA Tile

AI 导读

NVIDIA 在 CUDA Tile（cuTile）中实现了 Mamba-2 的高性能前向传播内核。cuTile 是基于 Python 的 DSL 与 GPU 并行编程模型，将线程级和 warp 级调度抽象为以 tile 为中心的高层原语，由编译器自动完成优化。开发者因此可用 Python 编写 GPU 内核并调用先进硬件能力。

正文 · 原文

$Decorative math image.$ NVIDIA CUDA Tile (cuTile) is a Python-based DSL and parallel programming model for NVIDIA GPUs that automatically leverages advanced hardware capabilities to... $Decorative math image.$

NVIDIA CUDA Tile (cuTile) is a Python-based DSL and parallel programming model for NVIDIA GPUs that automatically leverages advanced hardware capabilities to deliver peak performance for tile-level GPU programming. By abstracting thread- and warp-level scheduling behind high-level, tile-centric primitives, cuTile enables you to write GPU kernels in Python while compiler optimizations automatically…

Source

来源：NVIDIA Technical Blog · developer.nvidia.com