distil-labs/distil-gemma3-270m-SHELLper
distil-labs/distil-gemma3-270m-SHELLper is a 270 million parameter Gemma 3 model fine-tuned by Distil Labs for multi-turn bash function calling. Utilizing knowledge distillation from a 120B parameter teacher model, it achieves 96% tool-call accuracy on its test set. This compact model is optimized for local execution and excels at translating natural language requests into bash commands for tasks like file system navigation and manipulation.
Loading preview...
Model Overview
distil-gemma3-270m-SHELLper is a specialized 270 million parameter model from the Gemma 3 family, developed by Distil Labs. It is specifically fine-tuned for multi-turn bash function calling, enabling natural language interaction with command-line tools. The model was created using knowledge distillation from a much larger GPT OSS 120B teacher model, allowing it to achieve high performance in a significantly smaller footprint.
Key Capabilities
- High Accuracy: Achieves 96% tool-call accuracy on multi-turn bash command conversations, closely matching the performance of its 120B teacher model.
- Efficient Size: With only 270 million parameters, it is designed to run efficiently on local machines.
- Bash Command Support: Supports 20 common bash commands, including
ls,cd,cp,mv,rm,mkdir, andgrep. - Multi-turn Conversations: Capable of handling sequential commands and context within a conversation, though optimized for single tool calls per turn and best with up to 5 turns.
Use Cases
- Natural Language Interfaces: Ideal for creating intuitive interfaces to file systems and command-line environments.
- Command-Line Automation: Automating repetitive bash tasks through natural language prompts.
- Developer Productivity: Enhancing developer workflows with AI-powered command assistance.
- Educational Tools: Assisting in learning and practicing bash commands.
- Local AI Assistants: Deploying privacy-preserving AI assistants for system interaction.