Model Overview

Llama-3.1-Centaur-70B, developed by Marcel Binz, is a specialized foundation model built on the Llama 3.1 architecture. Its core purpose is to act as a "cognition model" capable of predicting and simulating human behavior within behavioral experiments described using natural language. This model is distinct in its focus on cognitive modeling rather than general-purpose language generation.

Key Characteristics

• Human Cognition Simulation: Designed specifically to model and predict human choices and behaviors in experimental settings.
• Specialized Prompting: Requires prompts to encapsulate human choices using "<<" and ">>" tokens for optimal performance.
• Research-Oriented: Primarily intended for academic and research applications in cognitive science and AI.
• Computational Requirements: Utilizes 70 billion parameters and requires significant computational resources, specifically 2 or more 80GB GPUs (NVIDIA Ampere or newer) and at least 150GB of disk space for deployment. A low-rank adapter is available for single 80GB GPU use with Unsloth.

Use Cases

• Predicting Human Behavior: Ideal for researchers studying human decision-making and cognitive processes.
• Simulating Behavioral Experiments: Can be used to run simulations of psychological experiments to understand potential human responses.
• Cognitive Science Research: A valuable tool for advancing research in the field of human cognition and AI's role in understanding it.

For more detailed information, refer to the project documentation and the associated research paper.