spade-rl/SPADE-Qwen3-4B-Games
SPADE-Qwen3-4B-Games is a 4 billion parameter model developed by spade-rl, fine-tuned from Qwen3-4B-Instruct-2507. This model is specifically designed for game environments, acting as both an Environment Designer and a Reasoning Agent. It excels at generating executable game environments and solving them, with a curriculum that adapts to the agent's current capabilities.
Loading preview...
SPADE-Qwen3-4B-Games Overview
SPADE-Qwen3-4B-Games is a 4 billion parameter model developed by spade-rl, built upon the Qwen/Qwen3-4B-Instruct-2507 base model. This model is uniquely designed for the games setting, operating in a dual role as an Environment Designer and a Reasoning Agent. The core innovation lies in its self-improving curriculum: the Designer is rewarded for creating environments that push the boundaries of what the Agent can currently solve, ensuring continuous progression and adaptation.
Key Capabilities
- Adaptive Environment Generation: The model dynamically creates executable game environments, with the difficulty scaling based on the Agent's performance.
- Problem Solving: Functions as a Reasoning Agent to solve the environments it or other Designers create.
- Curriculum Learning: Implements a policy-driven curriculum where environment complexity evolves with the Agent's capabilities, as detailed in the SPADE paper.
- Extended Context Length: Supports a context length of 262,144 tokens, enabling complex game scenarios and interactions.
Good For
- AI Research in Games: Ideal for researchers exploring adaptive curriculum learning, environment generation, and agent reasoning within game settings.
- Game Development: Potentially useful for procedural content generation or dynamic difficulty adjustment in games.
- Reinforcement Learning: Provides a framework for training agents in progressively challenging environments.