Components

OMRP is built upon several key components that work together to provide efficient, secure, and decentralized AI model discovery and interaction. This section provides an in-depth look at each of these core components.

1. Discovery Service

The Discovery Service is the central component responsible for matching user prompts with the most suitable AI models. It implements the core logic of the Optimal Model Retrieval Protocol.

Key features:

  • Receives and processes encrypted user prompts

  • Implements the Generative Prompt Search Model for optimal model selection

  • Coordinates with other OMRP components to find and rank suitable models

  • Supports both free and paid model suggestions

  • Returns selected model identifiers to the calling system

Key processes:

  1. Query Embedding Generation: Converts user prompts into vector representations.

  2. Model Embedding Retrieval: Fetches relevant model embeddings from the ModelDB.

  3. k-NN Search: Performs a k-Nearest Neighbors search using cosine similarity to find the best matching models.

  4. Coordinate with the Discovery Contract for final model selection

2. Prompt Categorization Module

This module categorizes incoming prompts into predefined categories to facilitate efficient model selection.

Key features:

  • Analyzes prompt content and context

  • Assigns prompts to relevant categories

  • Enhances the accuracy of model matching

  • Continuously learns and updates categorization based on user feedback and system performance

3. Embedding Service

The Embedding Service is responsible for generating vector representations (embeddings) for both user queries and AI models. These embeddings are crucial for the efficient operation of the discovery process.

Key features:

  • Utilizes state-of-the-art language models for embedding generation

  • Supports multi-modal embeddings (text, image, audio) if required

  • Implements caching mechanisms for improved performance

  • Provides a standardized interface for embedding generation

4. Model Database

The Model Database serves as a repository for AI model information.

Key features:

  • Stores model capabilities, specifications, and performance metrics

  • Maintains pre-computed model embeddings

  • Implements a Model Profiling and Scoring System

  • Provides efficient querying mechanisms for relevant model retrieval

  • Updates model information and embeddings as new models are added or existing ones are updated

5. Discovery Contract

The Discovery Contract is a smart contract that implements core OMRP functionalities on-chain:

Key features:

  • Ensures transparency and immutability of the model selection process

  • Interacts with the Discovery Service to provide on-chain verification of model selection

This contract serves as the on-chain component of the Discovery Service, ensuring transparency and immutability of the model selection process.

6. Integration Interfaces

While not core components of OMRP itself, the protocol provides integration interfaces for seamless interaction with external systems.

Key interfaces:

  • Input Interface: Receives user prompts and associated metadata

  • Output Interface: Returns selected model identifiers and relevant metadata

  • Configuration Interface: Allows adjustment of discovery parameters and algorithms

These core components work in concert to provide a robust and efficient system for AI model discovery. OMRP's modular design allows for easy updates and improvements to individual components without disrupting the entire system.

The following sections will delve deeper into the workflows and processes that tie these components together, specifically in the context of model discovery and selection.

Last updated