FireFunction V2: Advanced Function Calling

FireFunction V2, developed by Fireworks AI, is a 70 billion parameter model based on the Llama 3 architecture, specifically engineered for high-performance function calling. It represents a significant advancement over its predecessor, FireFunction v1, with substantial quality improvements across various metrics.

Key Capabilities

• Parallel Function Calling: Unlike FireFunction v1, V2 supports parallel execution of multiple functions, enhancing efficiency for complex tasks.
• Competitive Performance: Benchmarks show FireFunction V2 scoring 0.81 on public function-calling evaluations, competitive with GPT-4o's 0.80. It also retains strong instruction-following, scoring 0.84 on MT-bench.
• Broad Use Cases: Optimized for general instruction following, multi-turn chat with function calls, single and parallel function calling, and structured information extraction.
• Extensive Function Support: Capable of handling up to 20 function specifications simultaneously within its 8k context window.

Ideal Use Cases

• Tool-use applications: Integrating external tools and APIs via function calls.
• Complex conversational agents: Managing multi-turn dialogues that require dynamic function invocation.
• Structured data extraction: Accurately extracting specific information from unstructured text using defined schemas.

This model is not optimized for scenarios requiring 100+ function specifications or nested function calling.