Latte

A mini deeplearning inference framework inspired by Caffe

Highlight

Compared with Caffe, Latte has the following features:

• Only support forward computation for inference
• Using modern CMake based build system
• Managing third-party libraries using CMake
• Simplify dependencies, remove boost, etc.

Dependencies

• GCC
• CUDA
• cuDNN
• CMake >= 3.18

Build from source

1. Build docker development image

docker build -t latte:dev -f docker/Dockerfile .

2. Build

cmake -S . -B build -G Ninja
cmake --build build