Skip to content

ai-in-pm/Titans---Learning-to-Memorize-at-Test-Time

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Titans Demonstration Platform

This platform implements seven AI agents demonstrating key concepts from the paper "Titans: Learning to Memorize at Test Time". Each agent specializes in a different aspect of the architecture and works collaboratively to provide a comprehensive understanding.

The development of the Github Repository was inspired by the "Titans: Learning to Memorize at Test Time" paper. To read the full paper, visit https://arxiv.org/pdf/2501.00663v1

Agents

  1. OpenAI Agent (Neural Memory Module)

    • Demonstrates memory mechanisms
    • Real-time decay simulation
    • Interactive memory retrieval
  2. Anthropic Agent (Memory as Context)

    • Historical context integration
    • Language modeling demonstrations
  3. Mistral Agent (Memory as Gate)

    • Short-term/long-term memory gating
    • Efficiency demonstrations
  4. Groq Agent (Memory as Layer)

    • Neural network layer integration
    • Architecture comparisons
  5. Gemini Agent (Experimental Validation)

    • Scalability demonstrations
    • Information retrieval tasks
  6. Cohere Agent (Innovations)

    • Memory design principles
    • Real-world applications
  7. Emergence Agent (Analysis)

    • Architecture analysis
    • Scalability discussions

Setup

  1. Install dependencies:

    pip install -r requirements.txt
  2. Set up environment variables in .env:

    • OPENAI_API_KEY
    • ANTHROPIC_API_KEY
    • MISTRAL_API_KEY
    • GROQ_API_KEY
    • GOOGLE_API_KEY
    • COHERE_API_KEY
    • EMERGENCE_API_KEY
  3. Run the platform:

    python main.py

Usage

  1. Access the web interface at http://localhost:8000
  2. View demonstrations via /demonstrate endpoint
  3. Interact with agents through WebSocket connections
  4. View real-time visualizations and metrics

Features

  • Real-time demonstrations
  • Interactive agent communication
  • Dynamic visualizations
  • Performance metrics
  • Collaborative analysis

Architecture

The platform uses a modular architecture where each agent implements the base TitansAgent interface. The main orchestrator manages agent interactions and provides a unified API for demonstrations and user interactions.

About

Titans - Learning to Memorize at Test Time

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published