Hermes 2 Pro, an advanced iteration of the Nous Hermes 2 model, introduces significant enhancements in AI capabilities, particularly in function calling and JSON structured outputs. Developed through a collaboration between Nous Research, @interstellarninja, and Fireworks.AI, this model leverages an updated OpenHermes 2.5 Dataset alongside a novel Function Calling and JSON Mode dataset. It showcases exceptional performance, with a 90% score in function calling evaluations and an 84% in structured JSON output assessments. Hermes 2 Pro employs a unique system prompt and multi-turn function calling structure, facilitated by new tokens designed to streamline parsing while streaming tokens. The model supports ChatML for prompt formatting, enabling more structured and multi-turn chat dialogues, and is compatible with OpenAI’s API.
For function calling, Hermes 2 Pro uses specific system prompts and structures, allowing for precise and reliable function execution. This model is also adept at generating structured JSON outputs, following a defined JSON schema. Benchmarks across various tasks demonstrate its broad applicability and effectiveness in handling complex queries and tasks. Example inference code is provided for utilizing the model, highlighting its integration with HuggingFace Transformers and its efficiency in generating responses. For those interested in function calling capabilities, detailed documentation and code examples are available on the Nous Research GitHub repository. Hermes 2 Pro represents a significant step forward in the development of AI models capable of sophisticated task execution and interaction.
Read more…