🌐 Lite Translator for Mac (TranslateGemma)

中文说明可点击下方：

• 简体中文

Experience AI translation that feels native to macOS.

This is a high-performance local translation tool built specifically for Apple Silicon Macs (M1/M2/M3/M4). Powered by Apple's MLX framework, it runs Large Language Models (LLMs) entirely on your device.

The best part? It integrates seamlessy into your system via macOS Shortcuts. You can translate text with a Right-Click or a Keyboard Shortcut from any app—just like a built-in system feature!

No API keys. No internet required. 100% Private.

Demo

Demo.mp4

📸 Screenshots

🤖 About the Model & Performance

This application uses translategemma-12b-it-4bit, a specialized fine-tune of Google's Gemma model optimized for translation tasks.

• Performance: On Apple Silicon, the MLX framework allows this 12B parameter model to run with exceptional speed (often exceeding reading speed).
• Hardware Requirements:
  • Chip: Apple M1/M2/M3/M4 (Pro/Max/Ultra recommended for best speeds, but base chips work).
  • RAM: 16GB Unified Memory is recommended.
  • Note: The 4-bit quantized model requires approximately 8-9GB of RAM. It may run on 8GB machines if no other heavy apps are open.

✨ Features

• 🚀 Instant & Accurate: Click and use immediately. Delivers high-quality translations without needing an internet connection (once the model is downloaded).
• 🎨 Native Mac UI:
  • System Integration: Supports macOS Menu Bar and Dock operations.
  • Theme Support: Automatically adapts to Light Mode and Dark Mode.
  • Beautiful Icon: Features a native-style icon (Source: macOSicons.com, you can customize it easily).
• 🧠 Context Aware:
  • Auto-Dictionary Mode: Detects single words/phrases and automatically provides definitions, phonetics (IPA), and examples.
  • Smart Style Switching: Choose between Default, Academic, Casual, Web Chat, and more.
• 🔄 Smart Swap: Swapping languages automatically detects the original text language to reset the target direction.

🌍 Supported Languages

The app comes pre-configured with the following languages:

• English
• Simplified Chinese (简体中文)
• Traditional Chinese (繁體中文)
• Japanese (日本語)
• Korean (한국어)
• French (Français)
• German (Deutsch)
• Italian (Italiano)
• Spanish (Español)
• Russian (Русский)
• Portuguese (Português)
• Arabic (العربية)
• Hindi (हिन्दी)
• Maltese (Malti)

📝 Note: The underlying translategemma model actually supports 55 languages. If your desired language is not listed, you can easily add it by modifying the self.languages list in the source code.

📥 Download & Setup

Before configuring or running the app, you need to download the source code and install the required dependencies.

1. Get the Code

You can clone the repository via Git or download the ZIP file directly.

• Method A: Git Clone (Recommended) Open your Terminal and run:

  git clone https://github.com/Jingyuan-Zheng/Mac-Lite-Translator_TranslateGemma.git
  cd Mac-Lite-Translator_TranslateGemma

• Method B: Download ZIP Click the green Code button at the top right of this page -> Select Download ZIP. Unzip the file and navigate to the folder in your Terminal:

  cd /path/to/your/downloaded/Mac-Lite-Translator_TranslateGemma

2. Install Dependencies

Make sure you have Python installed (Python 3.10+ is recommended). Run the following command to install all required libraries (including MLX) listed in requirements.txt:

pip install -r requirements.txt

⚙️ Configuration (Language Setup)

Before running the app, open translator.py in any text editor (like VS Code, TextEdit, or Sublime Text) to set your preferred languages.

Look for the USER CONFIGURATION section near the top:

# ================= USER CONFIGURATION =================
USER_NATIVE_LANG = "English"           # Your Native Language
USER_PRIMARY_FOREIGN_LANG = "Español"  # Your Primary Target Language
# ======================================================

• Why do this? This determines the default translation direction. If you copy text in your native language, it auto-translates to the foreign one, and vice versa.
• Note: Ensure the language names match exactly with the self.languages list further down in the code.

🚀 Installation & Usage Methods

You can run this tool in three ways. Choose the one that fits your workflow.

Method 1: macOS Shortcuts (Recommended - Native Experience)

[🤔 If you are lazy, you can just download the shortcut on this repository and change the path in the shortcut.]

Integrate the translator directly into macOS so you can right-click any text to translate it.

1. Find your Python Path:

   • Open Terminal.
   • If you use Anaconda, type: which python (make sure your conda env is active). It usually looks like /opt/anaconda3/bin/python.
   • If you use a Virtual Environment, use the absolute path to the python executable inside it (e.g., /Users/name/my_project/.venv/bin/python).
   • Copy this path, you will need it.

2. Create the Shortcut:

   • Open the Shortcuts app.

   • Click + to create a new shortcut.

   • Right Sidebar (Info/i):

     • Check "Show in Share Sheet".
     • Check "Use as Quick Action".
     • Ensure "Services Menu" and "Finder" are selected.
     • (Optional) Add a keyboard shortcut (e.g., control+T).

   • The Workflow Actions:

     • Input: Set to "Receive Text from Quick Actions, Share Sheet".
     • Action: Drag in "Run Shell Script".
     • Shell: Select /bin/zsh.
     • Pass Input: Select "as arguments" (Crucial!).
     • Script: Enter the command below (replace paths with yours):

     # Template
     /FULL/PATH/TO/YOUR/python '/FULL/PATH/TO/translator.py' "$1"
     
     # Example (Anaconda User):
     /opt/anaconda3/bin/python '/Users/yourname/Documents/translator.py' "$1"

3. How to use:

   • Highlight text in any app (Chrome, PDF, Notes, etc.).
   • Right-Click -> Services -> Translate Text.
   • (Or use the keyboard shortcut you defined).

Method 2: Create a Standalone App (via Automator)

If you prefer a clickable app icon in your Dock or Launchpad.

1. Open Automator on your Mac.
2. Choose "Application" as the document type.
3. Search for "Run Shell Script" and double-click to add it.
4. Set Pass input to "as arguments".
5. Paste the same command as above:

   /opt/anaconda3/bin/python '/Users/yourname/Documents/translator.py'

   (Note: You don't need "$1" here if you just want to open the empty window).
6. Save it as AI Translator.app to your Applications folder.

Method 3: Command Line (Terminal)

For developers or debugging.

• Direct Run:

  # Ensure you are in the correct directory or use full paths
  python translator.py

• Pipe Text (Instant Translate): Great for scripting.

  echo "This is a test sentence." | python translator.py

• With Arguments:

  python translator.py "Hello world"

❓ F.A.Q.

Q: Does it support Multimodal input (Images)?

A: No. While Google's original Gemma model supports multimodal input, the current MLX version on HuggingFace does not. Detail: I tested a custom multimodal implementation, but it was excluded because it lacks streaming support and is significantly slower than the text-only version.

Q: Can I use Ollama instead of MLX?

A: Technically yes, but not in this repo. On macOS, MLX is significantly faster and more memory-efficient than Ollama due to direct optimization for Apple Silicon.

Q: Does this work on Windows or Linux?

A: Not out of the box. The UI is cross-platform (tkinter), but you would need to remove the macOS-specific code (AppKit, Dock icons) and replace the mlx-lm backend with ollama or transformers.