Study note

Get Started With Deep Learning Performance

View original Back to list

Properties

Type: Blogs
Status: 待读
Domain: AI / ML
Category: 深度学习、自动微分与实践
Source: docs.nvidia.com
Vault note: library/articles/ai_ml/Get-Started-With-Deep-Learning-Performance-cfa6139fd68064f9.md

解释矩阵乘为何是深度学习性能的核心，并给出 Tensor Core 对齐、batch size、矩阵维度和算子参数选择的实用建议，帮助判断 math-bound 与 memory-bound 工作负载。

适合紧接 GPU Performance Background 阅读，把 Roofline 和算术强度进一步落实到 GEMM、卷积及训练参数选择上。

NVIDIA 的深度学习 GPU 利用率入门指南。

Differentiable Programming from Scratch – Max Slater – Computer Graphics, Programming, and Math Blogs · 深度学习、自动微分与实践
Hands-On Modern RL Blogs · 深度学习、自动微分与实践
How to Build a Diffusion Language Model Blogs · 深度学习、自动微分与实践
MiniTorch Blogs · 深度学习、自动微分与实践
TensorTonic | Learn ML through code Blogs · 深度学习、自动微分与实践
Using group theory to explore the space of positional encodings for attention Blogs · 深度学习、自动微分与实践

Loading comments...