Seq2Seq: Train a German-to-English translation model

Understand the project

Practical overview

Use it when you want to explore or train a small neural machine translation model in code, or need a modular baseline to adapt to paired-text data.

Developers and learners studying a compact, readable PyTorch translation model.
Practitioners seeking a modular encoder-decoder starting point for sequence-to-sequence work.

Before you choose

The documented workflow needs Python 3.9+, PyTorch 2.0+, Hugging Face Datasets and spaCy 3.7+.
The README describes automatic device selection across CUDA, Apple MPS and CPU.
Changing the language pair can require dataset preprocessing, tokenization, vocabulary and spaCy model adjustments.
The repository specifies an MIT license.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install the documented dependencies and start training the German-to-English model.

Install the dependencies with the command below.
Documented command
```
pip install -r requirements.txt
```
Download the German and English spaCy models with the command below and the command below.
Documented command
```
python -m spacy download de_core_news_sm
```
Documented command
```
python -m spacy download en_core_web_sm
```
Start training with the command below.
Documented command
```
python train.py -epochs 30 -batch_size 32 -lr 3e-4
```

The documented training workflow starts a German-to-English model run using Multi30k data.

For CPU smoke runs, the README suggests smaller -hidden_size and -embed_size values.
The documented dataset and language models are German-to-English; other language pairs may require changes to preprocessing, tokenization, vocabulary and spaCy models.

Evidence and transparency

Source record details

Source availability: Active
Sources checked: 7 November 2026
Original website title: keon/seq2seq · Official homepage title · retained in its source language