If you are content using one of the SUMMA MT models, the easiest way is to just pull the docker image, e.g.
docker pull summaplatform/mt-engine-de-en:latest
The following engines are available:
- summaplatform/mt-engine-de-en:latest
- summaplatform/mt-engine-es-en:latest
- summaplatform/mt-engine-fa-en:latest
- summaplatform/mt-engine-lv-en:latest
- summaplatform/mt-engine-pt-en:latest
- summaplatform/mt-engine-ru-en:latest
- summaplatform/mt-engine-uk-en:latest
The image currently provides a RabbitMQ worker for use within the SUMMA platform and a batch translation script for testing. For the latter, use, e.g.
cat source.txt | docker run --rm -i summaplatform/mt-engine-de-en ./translate.py --cpu-threads=2 [-v]
By default, the image uses as many threads as the host has cpus, but models are loaded sequentially, so using more threads increases the startup time.
To use your own model, you can map it into the container running the MT engine:
cat source.txt \
| docker run --rm -i -v /path/to/my/model:/model:ro summa-platform/mt-engine \
./translate.py --cpu-threads=2 [-v]
The Docker image build is staged
For the MT engine image
git clone https:github.com/summa-platform/summa-mt.git
cd summa-mt
make image/mt-build-environment
make image/mt-marian-compiled
make image/mt-basic-engine
make image/mt-engine
To download all SUMMA models:
make models
To build all model images
make model-images
To build a single model image
make image/mt-model-${L}-en
Where ${L} is one of de, es, fa, lv, pt, ru, uk.
To build all engine images
make engine-images
To build a single engine image:
make image/mt-engine-${L}-en