# nvaitc-toolkit

**Repository Path**: mirrors_NVIDIA/nvaitc-toolkit

## Basic Information

- **Project Name**: nvaitc-toolkit
- **Description**: Open source code base to showcase interoperability of CUDA-X AI software stack in multi-GPU environments and thus provide researchers a reference framework to build new projects on.
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-11-20
- **Last Updated**: 2026-03-21

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

**NVIDIA AI Technology Centre (NVAITC)**
===============

# Toolkit 

### Introduction

Python codebase to showcase the interoperability of CUDA-X AI software stack in multi-GPU environments. The goal of this project is to provide researchers a reference framework to build new projects on. It requests the availability of ImageNet to demonstrate how to train a network (ResNet[18/50/101]) against a well known dataset. This codebase served as the underlying playground for the Oct 2020 NVAITC Webinar Series on Deep Learning available as a [YouTube playlist](https://www.youtube.com/watch?v=f-CvtA2nuN8&list=PL5B692fm6--sJLzBmCpUSpP36xUWwuO8c).

### Clone repo

```
git clone -b toolkit --single-branch https://github.com/nvidia/nvaitc-toolkit.git toolkit
```

### Getting Started

Please find details and installation instructions in [README.md](https://github.com/NVIDIA/nvaitc-toolkit/blob/toolkit/README.md).


# cuAugment

### Introduction

cuAugment is a CUDA-accelerated 1D/2D/3D/4D augmenter library that utilizes a just-in-time compiler to transform a cascade of coordinate transformation into a single monolithic kernel to avoid unnecessary accesses to global memory.

### Clone repo

```
git clone -b cuaugment --single-branch https://github.com/nvidia/nvaitc-toolkit.git cuaugment
```

### Getting Started

Please find details and installation instructions in [README.md](https://github.com/NVIDIA/nvaitc-toolkit/blob/cuaugment/README.md).

# Build Docker container from scratch 

The [Dockerfile](https://github.com/NVIDIA/nvaitc-toolkit/Dockerfile) available within this repository allows you build a new [docker](http://www.docker.com) container from scratch pulling both branches and installing packages needed to execute the code. 

Before executing the docker build command please [download](https://developer.nvidia.com/nccl) the NCCL package from the NVIDIA Developer zone and edit the Dockerfile accordingly.



```
git clone https://github.com/nvidia/nvaitc-toolkit.git
cd nvaitc-toolkit
docker build .
```