cloudsurf-software/CloudSurf-4B-FC
CloudSurf-4B-FC is a 7.9 billion parameter function-calling and tool-use model developed by CloudSurf Software, based on the Gemma-4 E4B architecture. It is QLoRA-tuned to excel at agentic tool use, achieving a 55.73 BFCL V4 FULL score, significantly outperforming stock Gemma-4-E4B-it and exceeding the small-model class bar. This model is optimized for robust function calling and integration into agentic workflows.
Loading preview...
CloudSurf-4B-FC: Function-Calling and Tool-Use Model
CloudSurf-4B-FC is a 7.9 billion parameter model from CloudSurf Software, QLoRA-tuned from the Gemma-4 E4B architecture specifically for function calling and agentic tool use. It demonstrates strong performance on the BFCL V4 FULL benchmark, scoring 55.73 (3-seed mean), which is a substantial improvement over the stock Gemma-4-E4B-it (34.81) and surpasses other small-model class benchmarks like Nanbeige4-3B (51.40).
Key Capabilities
- Superior Function Calling: Achieves high scores in function calling and tool-use scenarios, particularly in multi-turn interactions and web-based tasks.
- Agentic Tool Use: Optimized for integrating with external tools and APIs, with trained environmental API namespace competence.
- Robust Performance: Consistently exceeds established benchmarks for its model class, with audited clean data (0/255 eval-prompt 8-gram overlap).
- OpenAI-Compatible Serving: Includes a Python bridge for serving via an OpenAI-compatible
/v1/chat/completionsendpoint, facilitating easy integration into existing systems.
Good For
- Developers building AI agents requiring reliable function calling.
- Applications needing tool-use capabilities to interact with external systems.
- Use cases where multi-turn conversations and web research are critical for agent performance.
- Integration into OpenAI API-compatible environments for function calling.