FunctionGemma 270M (local function‑calling model)

A compact 2.6 billion parameter powerhouse optimized for precise tool use and local function calling.

FunctionGemma 270M leverages the Gemma 2 2B architecture to deliver high-performance tool integration on consumer-grade hardware. This model excels at converting natural language prompts into structured JSON arguments for external API calls, maintaining a small enough footprint to run entirely on-device. By utilizing the same sliding window attention and logit soft-capping as its larger 9B and 27B siblings, it provides reliable, low-latency reasoning for automated workflows and agentic systems without the overhead of cloud-based LLMs.

https://huggingface.co/google/gemma-2-2b-it