Gemma 3n-e4b-it Finetuned: Bertrand Duopoly and Smart Agent-Based Models

Developed by: carlosveras License: Apache-2.0 Finetuned from model: unsloth/gemma-3n-e4b-it-unsloth-bnb-4bit

Model Description

This model is a fine-tuned Gemma 3n architecture, optimized for operation as an agent within Smart Agent-Based Modeling (SABM) frameworks. Its performance is demonstrated in multi-round strategic interactions, specifically within the Bertrand Duopoly game.

The objective of this model is to provide a competent agent for simulations that demand computational efficiency. While large language models can offer sophisticated reasoning, their resource requirements can be substantial for agent-based simulations involving numerous agents. This fine-tuned Gemma 3n addresses this by offering a balance between effective decision-making and computational practicality, making it suitable for environments where many intelligent agents are required, including deployment on standard home computers.

Key Attributes:

Agent Functionality: Designed to perform as an agent making pricing decisions in a dynamic Bertrand Duopoly setting over multiple rounds.
Computational Efficiency: Built upon the Gemma 3n base, this model allows for the deployment of multiple agents without excessive computational load. This makes it viable for large-scale agent-based simulations and, importantly, enables execution on typical personal computer hardware.
Efficient Fine-Tuning: The model was fine-tuned using Unsloth and Huggingface's TRL library. Fine-tuning was conducted on a small dataset of previous SABM simulations generated using the Gemini 2.0 family of models.

Applications:

Smart Agent-Based Modeling (SABM): Suitable for integration into various SABM frameworks.
Economic Simulation: Applicable for simulating market dynamics and competitive behaviors in oligopolistic structures.
Local Deployment: Engineered to run effectively on home computers, lowering the barrier to entry for complex agent-based simulations and research.

This model contributes to the development of agent-based simulations that require both agent capability and efficient resource utilization, broadening accessibility for researchers and enthusiasts.