Skip to content

zengdard/Constellation-of-Models

Repository files navigation

Constellation of Models (CoM)

=====================================

Introduction


Welcome to the Constellation of Models (CoM) repository, where we introduce a groundbreaking approach to tackle complex, multi-modal tasks using artificial intelligence. The CoM system is designed to dynamically distribute tasks to a constellation of specialized, pre-trained models, each expert in its respective domain.

Table of Contents


  1. The Constellation of Models (CoM) Approach
  2. Router Nomenclature
  3. Models and Routers on Different Hardware Systems
  4. Security: Firewall at the Prompt Level
  5. Managing Multiple Models
  6. AI Model Types and Security Vulnerabilities
  7. Contact Information

The Constellation of Models (CoM) Approach


The CoM approach is a flexible and specialized solution for multi-model strategies. By setting a standard based on this approach, we can ensure consistency and efficiency in situations where multiple models must be utilized. The dynamic router serves as a skillful orchestrator, routing each query to the most capable model, optimizing performance and efficiency.

Router Nomenclature


The router nomenclature presented in this repository offers a clear and coherent naming convention for a wide range of AI models and tasks. The nomenclature adheres to a simple pattern, with router names prefixed by "CoM_" followed by a succinct descriptor of the model's function or task.

Models and Routers on Different Hardware Systems


The CoM approach accommodates the distribution of models across multiple servers, harnessing unique computational power and specialized capabilities. This distributed architecture allows for scalability, efficiency, and enhanced performance in AI systems.

Security: Firewall at the Prompt Level


To protect AI models from potential threats, a firewall at the prompt level is implemented, acting as a barrier that filters and monitors incoming requests to the AI models, ensuring that only legitimate and safe queries are processed.

Managing Multiple Models


As AI models proliferate, the CoM approach provides an intelligent management system that dynamically orchestrates tasks and data among models, ensuring optimal performance and resource allocation. Additionally, a framework is established to monitor, secure, and validate the behavior of each model within the CoM ecosystem.

AI Model Types and Security Vulnerabilities


The repository provides an overview of various AI model types and their associated security vulnerabilities, such as computer vision, natural language processing, audio processing, reinforcement learning, and tabular data processing models.

Contact Information


For any inquiries or additional information, please contact nielzac[@]proton.me.

image image

Credit

FineTune Code used : https://github.com/geronimi73/phi2-finetune/blob/main/nb_qlora.ipynb HTML RIKTIK

About

GitHub of Constellation of Models

Topics

Resources

Stars

Watchers

Forks

Languages