Firefunction
About
The Firefunction family of large language models is specifically designed for efficient function calling, enabling seamless interaction with external APIs and access to real-time data. The initial version, Firefunction v1, was built on the Mixtral 8x7B model and achieved GPT-4-level accuracy in practical scenarios, with the advantages of faster speeds and open-source accessibility. Its successor, Firefunction v2, utilizes the Llama 3 model to enhance its function-calling capabilities while maintaining robust conversational and instruction-following features. This version excels in multi-turn dialogues, parallel function calling, and consistently surpasses Llama 3 in function-calling tasks, delivering results akin to GPT-4o on public benchmarks. These models are tailored for real-world use, combining accuracy, speed, and cost effectiveness.