Jarvis (UI is coming...)

...sneak peek at the ui(still in dev) here

The J.A.R.V.I.S Large Language Model, 100% offline

We have used custom jarvis dataset and various other open datasets on the internet of jarvis's dialog with stark to fine-tune on top of Llama2-7b

Content

• Dataset here | dataset is being created progresively
• Model: here | Updated every week or so

Jarvis is being built with privacy in mind, everything runs locally. This fine-tuned model is better at responding like jarvis and producing response in the best jarvis-tone possible.

Star History

Current features

• Jarvis can tell you the current date and time just ask:) Try: What time is it? what is the date of today?
• Open your default browser and search for something Try: hey jarvis, help me search for the best african dish out there
• play youtube video by title Try: hey jarvis, play Girls like you by Maroon 5 on youtube
• You can ask jarvis what it sees(uses the webcam, only works in the project not the chat, you have to run the while project. NB: it uses llava(100% offline), so you have to download and install it with ollama command "ollama run llava" ) Try:"what is this?" "what are you looking at?", "tell me what you see", "describe this" or even "describe what you see"
• Search on wikipedia
• Send emails from contacts

Project structure

‼️Currently doing a major achitecture restructure(please check back later...)

Usage

Installation

I. You will need Ollama to download and install the model for use locally.

1. Download and install ollama here
2. Run the command:

ollama run fotiecodes/jarvis

This will install the model jarvis model locally. 3. From here you can already chat with jarvis from the command line by running the same command ollama run fotiecodes/jarvis or ollama run fotiecodes/jarvis:latest to run the lastest stable release.

II. After installing the model locally and started the ollama sever and can confirm it is working properly, clone this repositry and run the main.py file

🚨 Important

Please check the .env.example file and add the neccessary env variables before running.

That's it, you can start talking to Jarvis✨

Setup local dev

Please go to CONTRIBUTOR.md for more info.

To-Do:

• Prepare instruct dataset(WIP)
• Fine-tune llama2-7b for a costum model Llama-2-7b-chat-jarvis"(WIP)
• Build a proper voice clone of jarvis(In progress): currently having difficulties finding clean voice samples of jarvis on the internet, so working with what i have, i guess won't be perfect.
• Use Whisper for voice to text input(if anyone has anything better, please suggest as this is the best i found out there.)
• Deploy the voice model as an API for everyone to be able to use it locally(We can deploy on my vps i don't know, if the specs will even let us though, we'll see)

Installing TTS for local processing

for offline tts we switched from Kokoro to LuxTTS with custom module here JarvisLuxTTS, an amazing low latency tts(still working on this)

Note: after cloning the repo you need to type the following command to install all the necessary libries. this installs from the roor directory and sub directories

find . -name "requirements.txt" | while read req; do echo "Installing from $req..."; pip install -r "$req"; done

Note: the voice name is already set in the .env file(the actual jarvis voice will be coming soon) this one is closest we have at the moment to jarvis's voice.

Why Jarvis?

Well, simple. I have always wanted to have my very own Jarvis, now i'm not talking about a siri clone or google home assitant clone. I am talking about my very own Jarvis, talks like jarvis, response like jarvis, feels like jarvis in the most accurate way possible.

Acknowledgments

Special thanks to:

• ollama for the amazing project
• ggerganov the genius behind llama.cpp
• kokoro-onnx the tts that powers jarvis for offline inferencing
• Every contributor to this repo:)

Supporters ✨

https://github.com/thewh1teagle/kokoro-onnx/projects[]

• We're looking for Sponsors!

License 📜

This project is licensed under the MIT License - see the LICENSE file for details. 📄

Support 💬

If you have any questions, suggestions, or need assistance, please open an issue:)

Disclaimer: Jarvis can make mistakes. Consider checking important information.

* This project was inspired by "J.A.R.V.I.S" from the Marvel movie series