cmu-lti/osim-4b

OSim-4B: Human Behavior Simulation Model

cmu-lti/osim-4b is a 4 billion parameter language model specifically post-trained for simulating human behavior and user interactions. It is a sibling model to the 8 billion parameter cmu-lti/osim-8b-post.

Key Capabilities

• Human Behavior Simulation: Optimized to model and predict human actions and responses in various scenarios.
• User Simulation: Designed for applications requiring realistic user interaction and decision-making.
• Specialized Training: Built upon Qwen3-4B, it underwent a unique training regimen including OdysSim mid-training, reinforcement learning, and expert consolidation to enhance its simulation capabilities.

Good For

• Research in Human-AI Interaction: Ideal for studying and developing systems that interact with simulated human users.
• Virtual Agent Development: Useful for creating agents that mimic human behavior patterns.
• Simulation Environments: Applicable in scenarios where realistic human or user responses are critical for testing and development.

For more technical details, refer to the OdysSim paper on building foundation models for human behavior simulation.