Getting Started

Prerequisites

• Python 3.10+
• Ollama installed and running locally
• GPU with sufficient VRAM for your chosen model

Install

pip install polyglot-gpu

Choose a model

polyglot-gpu uses TranslateGemma via Ollama. Three sizes are available:

Model	VRAM	Speed	Quality
translategemma:4b	3.3 GB	Fast	Good
translategemma:12b	8.1 GB	Balanced	Recommended
translategemma:27b	17 GB	Slow	Best

The model is pulled automatically on first use. To pull it manually:

ollama pull translategemma:12b

First translation

import asyncio
from pypolyglot import translate

async def main():
    result = await translate("Hello world", "en", "ja")
    print(result.translation)

asyncio.run(main())

polyglot-gpu will auto-start Ollama if it’s not running and auto-pull the model if it’s not installed. On first run, expect a short delay while the model downloads.

Verify setup

Use the check_status tool to confirm everything is working:

from pypolyglot.server import check_status

# Returns Ollama status and installed models
status = await check_status()

Next steps

• Library Usage — Translate text, markdown, and multiple languages
• MCP Server — Use with Claude Code
• Configuration — Tune model and concurrency