Ricardo-H/BehR-WorldModel-Webshop-Llama3.1-8B
Ricardo-H/BehR-WorldModel-Webshop-Llama3.1-8B is an 8 billion parameter Llama-3.1-based text-based world model specifically designed for the WebShop environment. This model predicts the next environment observation given an agent's interaction history and next action, enabling it to act as a text simulator for agent rollouts. Its primary differentiation lies in its focus on behavior consistency in text-based world models, as detailed in its associated research paper. It is intended for research in world models and agent simulation within text-based environments.
Loading preview...
Overview
Ricardo-H/BehR-WorldModel-Webshop-Llama3.1-8B is an 8 billion parameter model built on the Llama-3.1 architecture, specifically developed as a behavior-consistent text-based world model for the WebShop environment. It is designed to predict the next environment observation based on an agent's interaction history and its subsequent action, effectively functioning as a text simulator for agent rollouts.
Key Capabilities
- Environment Simulation: Acts as a text simulator for agent rollouts within the WebShop environment.
- Behavior Consistency: Focuses on maintaining behavior consistency in text-based world models, as explored in its accompanying research paper: Beyond State Consistency: Behavior Consistency in Text-Based World Models.
- Observation Prediction: Given an agent's interaction history and next action, it predicts the subsequent environment observation.
Intended Use and Limitations
- Good for: Research on world models and agent simulation in text-based environments, particularly within the WebShop domain.
- Limitations: Evaluated exclusively on WebShop; its behavior outside this distribution is unvalidated. Predicted observations, while plausible, may not always be factually accurate and should not be treated as ground truth. The model inherits biases and failure modes from its base models, Llama 3.1 8B and the underlying world model.