These templates provide a great starting point for building agents with the HuggingFace ecosystem. Check out the smolagents documentation to learn more.
  • SmolCodingAgent: A text-based agent that uses a code-generating model to reason about the game and execute actions as Python code.
  • SmolVisionAgent: A multimodal agent that processes game frames as images, allowing it to “see” the game state.
To run these agents, use the following commands: 
# Run the text-based coding agent
uv run main.py --agent=smolcodingagent --game=ls20
`
# Run the vision-based agent
uv run main.py --agent=smolvisionagent --game=ls20
Source File