miniLernen

learn less is more

try to use less code to get insight of different frameworks.

Contents

Basic

Hello world

Parallel Computating

CPU

• neon(TODO)

GPU

CUDA

• CUDA kernel

• CUDA kernel in python

• CUDA kernel in tensorrt(TODO))

• opencl

• cublas

Neural Network

model architecture

transformer

• self attention
• multi-head attention
• cross attention

model deployment

Pytorch

jit(TODO)

torchscrpt(TODO)

onnx

• onnx export
• onnx graphsurgeon