EasyNMT provides a user-friendly Python interface for state-of-the-art Neural Machine Translation, supporting over 100 languages. It simplifies the integration of advanced translation models like Opus-MT and Facebook's mBART and M2M series, making high-quality machine translation accessible with minimal code.

How It Works

EasyNMT acts as a wrapper around pre-trained Transformer-based models. It handles automatic model downloading, language detection, and sentence splitting for document translation. The library supports multi-GPU and multi-process translation for improved performance and scalability.

Quick Start & Requirements

• Install via pip: pip install -U easynmt
• Requires PyTorch with GPU support for optimal performance.
• See EasyNMT Google Colab Example for a quick start.

Highlighted Details

• Supports 150+ languages via Opus-MT models.
• Offers automatic language detection for 170+ languages.
• Includes ready-to-use Docker images for REST API deployment.
• Provides wrappers for mBART50 and M2M_100 models from Facebook AI.

Maintenance & Community

• Developed by UKPLab at TU Darmstadt.
• Contact: info@nils-reimers.de.
• Issues can be reported via the GitHub repository.

Licensing & Compatibility

• The README does not explicitly state a license. The models themselves are subject to their original licenses (Helsinki-NLP, Facebook Research). Compatibility for commercial use or closed-source linking requires verification of underlying model licenses.

Limitations & Caveats

• Input length is limited by the underlying models (e.g., 512 word pieces for Opus-MT).
• The 12 billion parameter M2M model is not currently supported.
• Translation quality comparison between models is pending.