GitHub - microsoft/JARVIS

Description

JARVIS is a collaborative system that utilizes LLMs to connect with various AI models for complex tasks. The system, developed by Microsoft, aims to advance research in artificial general intelligence (AGI). It operates through a four-stage workflow: Task Planning, Model Selection, Task Execution, and Response Generation. The system uses ChatGPT to analyze user requests, select expert models from Hugging Face, execute tasks, and generate responses.

JARVIS offers several features, including support for OpenAI services on the Azure platform and the GPT-4 model. It provides CLI and web modes for interaction. The system supports different configurations, including a 'lite' version that doesn't require local model deployment, and a 'default' version with more extensive requirements. The project also includes a Gradio demo and web API access for intermediate results. The system's architecture allows for the integration of various models, enhancing its capabilities in task automation and response generation.

JARVIS is designed for researchers, developers, and anyone interested in exploring the capabilities of LLMs and AI task automation. The system's modular design allows for easy integration of new models and functionalities. The project provides detailed instructions for setup and usage, including server configuration and web application deployment. The project also provides example requests and responses to help users understand how to interact with the system. The project's documentation includes information on system requirements, quick start guides, and citation information for the related research papers.

JARVIS's value lies in its ability to streamline complex AI tasks by integrating LLMs with expert models. It provides a platform for researchers to experiment with different models and configurations. The system's flexibility and modular design make it a valuable tool for anyone interested in exploring the potential of LLMs in AI.

GitHub's Core Features

• Integrates LLMs with expert models

• Supports task planning and execution

• Utilizes ChatGPT for task analysis and response generation

• Offers CLI and web modes

• Supports OpenAI and GPT-4

• Provides a Gradio demo

• Offers a web API for intermediate results

• Includes a 'lite' configuration for easier setup

• Supports local and Hugging Face Inference Endpoints

• Facilitates AI task automation

• Modular design for easy integration of new models

Getting Started with GitHub

1. 1

   Clone: Clone the JARVIS repository from GitHub.

2. 2

   Install dependencies: Set up the environment and install necessary packages using conda and pip.

3. 3

   Configure: Replace API keys and tokens in the configuration files.

4. 4

   Download models: Download the required models using the provided script.

5. 5

   Execute: Run the server and the application in either server, CLI, or web mode.

6. 6

   Access: Access the services via the Web API or the Gradio demo.

7. 7

   Test: Test the system by sending requests and analyzing the responses.

8. 8

   Optimize: Explore different configurations and models to optimize performance.

GitHub's Use Cases

• Task Automation
• Model Integration
• AI Research
• Web API Access
• Gradio Demo
• CLI Interaction
• Model Evaluation

FAQ from GitHub

• What is JARVIS?

  JARVIS is a system developed by Microsoft to connect LLMs with the ML community, designed to explore artificial general intelligence (AGI).

• How does JARVIS work?

  JARVIS operates in four stages: Task Planning, Model Selection, Task Execution, and Response Generation, using ChatGPT as a controller and expert models as executors.

• What are the system requirements for JARVIS?

  The requirements vary depending on the configuration. The 'default' configuration requires Ubuntu 16.04 LTS, VRAM >= 24GB, RAM > 12GB, and disk space > 284GB. The 'lite' configuration has minimal requirements.

• How do I start using JARVIS?

  First, replace your OpenAI Key and Hugging Face Token in the configuration file. Then, run the setup commands for the server and the application.

• What modes are available for interacting with JARVIS?

  JARVIS offers CLI mode, server mode with a Web API, and a Gradio demo for interaction.

• Where can I find the Gradio demo?

  The Gradio demo is hosted on Hugging Face Space.

• What is the purpose of the configuration files?

  Configuration files, such as config.default.yaml and config.lite.yaml, specify the settings for the server, including model selection and inference endpoints.

• What is the difference between the 'default' and 'lite' configurations?

  The 'lite' configuration does not require any expert models to be downloaded and deployed locally, while the 'default' configuration requires more resources and allows for local model deployment.

• How can I cite JARVIS?

  You can cite the paper as provided in the README file, including the authors and publication details.

• What are the supported inference modes?

  The supported inference modes are local, huggingface, and hybrid.