Granite3.2-Gemma3 AI Model User Manual

1. Model Overview

This document serves as the official user manual for the Granite3.2-Gemma3 AI model, providing detailed information on its basic parameters, usage methods, supported scenarios, and precautions to help users quickly get started and use the model effectively.

Model Name: samuser3/granite3.2-gemma3

Model Author: samuser3

Core Capabilities: The model integrates multiple core functions, including vision recognition, tool calling, logical thinking, text generation, and code generation, capable of handling a variety of complex tasks.

Model Versions: The model provides two versions to meet different user needs and hardware configurations:

• 1B Version: File size is 523MB, context window is 40K, and only supports text input.

• 4B Version: File size is 3.3GB, context window is 128K, and supports both text and image input, with stronger multi-modal processing capabilities.

2. Supported Scenarios

The Granite3.2-Gemma3 AI model is widely applicable to various scenarios, including but not limited to:

• Intelligent Conversation & Question Answering: Conduct natural and smooth conversations with users, answer various questions, and provide professional and accurate responses.

• Code Generation & Optimization: Generate code snippets in multiple programming languages (such as Python, JavaScript), and assist in code debugging and optimization to improve development efficiency.

• Image Understanding & Analysis (4B Version Only): Recognize and analyze image content, extract key information, and complete image-related reasoning tasks.

• Logical Reasoning & Thinking Tasks: Complete complex logical reasoning, problem-solving, and thinking tasks, and provide clear reasoning processes.

• Tool Invocation & Extended Task Execution: Invoke related tools according to user needs to complete extended tasks and expand the model’s application scope.

• Long-Text Processing & Document Understanding: Process and analyze long texts, extract core information, and assist users in understanding and sorting out document content.

3. Quick Start via Ollama

The model supports quick startup and calling through the Ollama platform. Before use, please ensure that the Ollama service has been installed and started locally. The specific calling methods are as follows:

3.1 CLI Command

Use the command line to directly run the model, which is suitable for quick testing and simple interaction:

ollama run samuser3/granite3.2-gemma3

3.2 cURL Request

Call the model through the API interface using cURL, which is suitable for integration into scripts or third-party applications:

curl http://localhost:11434/api/chat \
  -d '{
    "model": "samuser3/granite3.2-gemma3",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

3.3 Python Call

Call the model using Python code, which is convenient for developers to integrate into Python projects:

from ollama import chat

response = chat(
    model='samuser3/granite3.2-gemma3',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

3.4 JavaScript Call

Call the model using JavaScript code, which is suitable for front-end or Node.js projects:

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'samuser3/granite3.2-gemma3',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

4. Integrated Application Launch

The model can be launched and used in third-party clients through the following commands, realizing rapid integration and application:

# Claude Code
ollama launch claude --model samuser3/granite3.2-gemma3

# Codex
ollama launch codex --model samuser3/granite3.2-gemma3

# OpenCode
ollama launch opencode --model samuser3/granite3.2-gemma3

# OpenClaw
ollama launch openclaw --model samuser3/granite3.2-gemma3

5. Version Parameter Comparison

The following table details the parameter differences between different versions of the model, helping users choose the appropriate version according to their needs:

6. Usage Notes

• Before using the model, please ensure that the Ollama service is running locally; otherwise, the model cannot be called normally.

• The 4B version supports image input and has a longer context window, which requires more system resources (such as memory and CPU). It is recommended to use it on devices with sufficient hardware configuration.

• The model has the capabilities of vision recognition, tool use, and deep logical thinking, and can be flexibly used according to different task requirements.

• The model is suitable for daily chat, code development, document analysis, multi-modal understanding and other scenarios, and is not suitable for high-risk, illegal or unethical tasks.

• The model is continuously updated and optimized by the author. Please pay attention to the latest update information to obtain better use experience.