https://github.com/sbstndb/neural_k

A simple Neural Network library using Kokkos enabling CUDA or OpenMP backend
https://github.com/sbstndb/neural_k

ai cuda kokkos library neural-network openmp

Last synced: 10 days ago
JSON representation

A simple Neural Network library using Kokkos enabling CUDA or OpenMP backend

• Host: GitHub
• URL: https://github.com/sbstndb/neural_k
• Owner: sbstndb
• Created: 2024-08-15T13:07:57.000Z (6 months ago)
• Default Branch: main
• Last Pushed: 2024-09-12T21:01:38.000Z (5 months ago)
• Last Synced: 2024-12-10T22:41:20.738Z (2 months ago)
• Topics: ai, cuda, kokkos, library, neural-network, openmp
• Language: C++
• Homepage:
• Size: 26.4 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Introduction

 Here is a simple Neural Network project. 

 The aim of this project is to have a first look at the Kokkos library and AI science.

 From the [Kokkos](https://github.com/kokkos) github repo : 

> The Kokkos C++ Performance Portability Ecosystem is a production level solution for writing modern C++ applications in a hardware agnostic way.

Thanks to Kokkos, you can compile this code for multiple backends like OpenMP, CUDA, HIP.

**Disclaimer:** Please note that this project is a work in progress and may contain errors or programming oversights due to its experimental nature. Your understanding and feedback are appreciated as we continue to develop and refine this code.

# Compilation

Compile the code with the following commands : 

```

git submodule update --init --recursive

mkdir build && cd build && cmake ..

make -j 

```

# Usage

*You can currently launch the executable named `nk` with the following command :

```

./nk

```

This will run the executable with the default parameters.

# Prerequisites

The code use Kokkos as a performance portability library. Then, you must have it to compile the project.

I suggest you to install it and read the documentation for further understanding. 

It is possible to easily install `Kokkos` through the HPC [`spack`](https://github.com/spack) package manager. I suggest you to create a new environment : 

```

spack env create kokkos

spack env activate kokkos

spack install kokkos // here you can specify your backend like OpenMP, pthread, CUDA, HIP, ... please read the doc

spack load kokkos

```

# Todo 

### To-Do List

- [x] Provide the MVP (Minimal Viable Product)

- [ ] Support flexible network architectures

- [ ] Enable GPU-accelerated Execution (CUDA/HIP)

- [ ] Design user-friendly API

- [ ] Add tests

- [ ] Add documentation

- [ ] Implement batch training strategies