Google has unveiled FunctionGemma, a variant of Google's Gemma line of AI's designed for specialised tasks and instructions, according to a blog post from the company.
FunctionGemma is an open model that can process human verbal commands into computer code that it can then execute automatically, without further input. Unlike Gemini, Function Gemma works offline, without the use of the internet.
It works as an AI agent manager for the user's personal device, such as their smartphone. The AI has been built with developers in mind. The model spins out structured code such as JSON or API calls, required to perform a task.
The developer can then connect the code to the app's functions in order to execute them.
"The model is supported by popular tools across the entire workflow: fine-tune with Hugging Face Transformers, Unsloth, Keras or NVIDIA NeMo and deploy using LiteRT-LM, vLLM, MLX, Llama.cpp, Ollama, Vertex AI or LM Studio," the blog post said.
FunctionGemma is designed for the execution of specific set of functions such as smart home control or media navigation, it is also a much simpler version of Gemini focusing on following commands rather than generating content.
It is also designed to be energy efficient as well as provide instantaneous results, running on a localised device, such as a smartphone. If users require the execution of complex generative tasks, the AI can connect to Gemini or Gemma 27B for additional assistance to save time and power.
Business owners and executives can use FunctionGemma to deploy a fleet of specialized task models that are inexpensive, easy to train, power saving, efficient and ensure privacy, as its offline nature ensures that data does not leave the device.
Ex-Nvidia Billionaire Unveils New AI Chips After China IPO Debut
