Functionary-small-v2.4: Advanced Function Calling Model
Functionary-small-v2.4, developed by MeetKai, is a 7 billion parameter language model optimized for interpreting and executing functions and plugins. It features a 4096-token context length and is designed to intelligently determine when to call functions, whether in parallel or serially, and to understand their outputs for grounded responses.
Key Capabilities
- Intelligent Parallel Tool Use: Capable of orchestrating multiple tool calls concurrently.
- Output Grounding: Analyzes function/tool outputs to provide relevant and accurate responses.
- Conditional Tool Use: Decides when to use tools and when to provide a standard chat response.
- Code Interpreter Support: Includes functionality for code interpretation.
- High Function Calling Accuracy: Achieves 83.07% accuracy on the SGD dataset for function calling (name & arguments), outperforming OpenAI's GPT-3.5-turbo-1106 and GPT-4-1106-preview on this specific metric.
Unique Features
This model utilizes a "v2PromptTemplate" that structures turns with from, recipient, and content portions, converting function definitions into TypeScript-like syntax within system prompts. This specialized prompting enhances its ability to handle complex function calls. MeetKai encourages running the model via their OpenAI-compatible vLLM server for optimal performance and ease of use.