This repository provides examples for running AI models and applications on NVIDIA Jetson devices with a single command.
This repo builds upon the work of the jetson-containers, which provides a modular container build system for various AI/ML packages on NVIDIA Jetson devices.
- π Easy Deployment: Deploy state-of-the-art AI models on Jetson devices in one line.
- π Versatile Examples: Supports text generation, image generation, vision transformers, computer vision and so on.
- β‘ Optimized for Jetson: Leverages Nvidia Jetson hardware for efficient performance.
To install the package, run:
pip3 install jetson-examples
Notes:
- Check here for more installation methods
- To upgrade to the latest version, use:
pip3 install jetson-examples --upgrade
.
To run and chat with LLaVA, execute:
reComputer run llava
Here are some examples that can be run:
Example | Type | Model/Data Size | Docker Image Size | Command |
---|---|---|---|---|
π llama3 | Text (LLM) | 4.9GB | 10.5GB | reComputer run llama3 |
π ollama | Inference Server | * | 10.5GB | reComputer run ollama |
LLaVA | Text + Vision (VLM) | 13GB | 14.4GB | reComputer run llava |
Live LLaVA | Text + Vision (VLM) | 13GB | 20.3GB | reComputer run live-llava |
stable-diffusion-webui | Image Generation | 3.97G | 7.3GB | reComputer run stable-diffusion-webui |
nanoowl | Vision Transformers(ViT) | 613MB | 15.1GB | reComputer run nanoowl |
nanodb | Vector Database | 76GB | 7.0GB | reComputer run nanodb |
whisper | Audio | 1.5GB | 6.0GB | reComputer run whisper |
π yolov8-rail-inspection | Computer Vision | 6M | 13.8GB | reComputer run yolov8-rail-inspection |
Note: You should have enough space to run example, like
LLaVA
, at least27.4GB
totally
More Examples can be found examples.md
Want to add your own example? Check out the development guide.
We welcome contributions to improve jetson-examples! If you have an example you'd like to share, please submit a pull request. Thank you to all of our contributors! π
- check disk space enough or not before run
- allow to setting some configs, such as
BASE_PATH
- detect host environment and install what we need
- support jetson-containers update
- all type jetson support checking list
- better table to show example's difference
- try jetpack 6.0
This project is licensed under the MIT License.