Ricardo-H/BehR-WorldModel-Webshop-Qwen2.5-7B

Overview

Ricardo-H/BehR-WorldModel-Webshop-Qwen2.5-7B is a 7.6 billion parameter Qwen2.5-based causal language model, building upon X1AOX1A/WorldModel-Webshop-Qwen2.5-7B. Its core function is to predict the next environment observation in the WebShop environment, given an agent's interaction history and its subsequent action. This model emphasizes behavior consistency in text-based world modeling, as detailed in the paper "Beyond State Consistency: Behavior Consistency in Text-Based World Models" (arXiv:2604.13824).

Key Capabilities

• Text-based World Simulation: Acts as a text simulator for agent rollouts within the WebShop environment.
• Observation Prediction: Predicts the next environment observation based on agent actions and interaction history.
• Behavior Consistency: Focuses on maintaining behavior consistency in its predictions, a distinction from models solely focused on state consistency.

Intended Use and Limitations

• Research Focus: Primarily intended for research in world models and agent simulation within text-based environments.
• WebShop Specific: Evaluation and validated behavior are limited to the WebShop environment; performance outside this distribution is unvalidated.
• Not Ground Truth: Predicted observations, while plausible, may not be factually correct and should not be treated as ground truth for the real environment.
• Inherited Biases: Inherits biases and potential failure modes from its base models, Qwen2.5-7B and X1AOX1A/WorldModel-Webshop-Qwen2.5-7B.