gemma

Explicit Vertex AI Text Generation Skill using Gemma Models (Supports Brand Guidelines & Variants)

SKILL.md

GitHub

Gemma - Explicit Vertex AI Model Garden Interaction

Overview

This skill executes chat/generation requests against custom deployed Gemma endpoints or MaaS instances in the bulachak Vertex AI project. It supports variant routing (pointing to specific deployed Endpoints) and optional brand guideline injection.

When to Use This Skill

Activate this skill when the user runs the /gemma command to generate text or interact with a Gemma variant.

Usage

Run the script using the python interpreter in the virtual environment.

Arguments:

  • --prompt: (Required) The base prompt for generation.
  • --variant: (Optional) The Gemma variant or Endpoint mapping (e.g., 4-pro). Defaults to the standard Gemma 4 MaaS endpoint.
  • --brand: (Optional) Path or name of the brand guideline file. If just a name (e.g., Baza), it checks 06_SOUL_KITCHEN/Brand_Books/.
  • --thinking: (Optional) Flag to enable the reasoning/thinking mode (if supported by the model).

Example:

python app/tools/model_garden/gemma_cli.py --prompt "Write a short blog post" --variant 4-pro --brand BazaCreative

Expected Output

The script will output the AI's response directly to the console. If a brand file is injected, it will confirm the injection prior to execution.