projectNMT

March 18, 2026

GitHub Repository

projectNMT focuses on practical neural machine translation for English to German/French using a research-style codebase based on fairseq components.

What it does

• Trains translation models using encoder-decoder architectures.
• Runs batch generation and interactive translation.
• Supports preprocessing/tokenization workflows (including SentencePiece utilities).
• Packages translation models for local service workflows.

How it is built

• Includes a full fairseq code tree with task/model/criterion registries.
• Uses dedicated scripts for training (train.py), generation (generate.py), and interactive decoding.
• Adds utility scripts for binarized dataset inspection and score analysis.
• Provides project-level instructions for deploying trained checkpoints for EN-DE and EN-FR use cases.

Tech stack

Python, PyTorch, fairseq, SentencePiece, sacrebleu and related MT tooling.