ucrel/grobid repository overview

⁠Grobid Docker

This README is taken from the repository that this docker image was created from: https://github.com/UCREL/grobid-docker⁠

Currently works with Grobid version:

• 0.6.1

Each docker tag is associated with the version of Grobid it was built with e.g. ucrel/grobid:0.6.1 means that it was built with version 0.6.1

This contains the Docker files to build the Grobid server⁠. To build the image that can run the Grobid server use the script:

bash make_docker.sh

This built image can be found on the UCREL Docker hub: https://hub.docker.com/repository/docker/ucrel/grobid⁠

For details on the script that builds the docker image see the section script in detail.⁠

For details on the Grobid server API: https://grobid.readthedocs.io/en/latest/Grobid-service/⁠

⁠How to run Grobid server with docker

The Grobid server uses two ports:

1. 8070 -- this is the main port which the server uses as an API to Grobid
2. 8071 -- this is the admin console.

To run with default settings:

docker run --rm --init -p 127.0.0.1:8070:8070 -p 127.0.0.1:8071:8071 ucrel/grobid:0.6.1

To run restricting the amount of overall memory Docker will use, in this case 5GB, and the amount of memory the pdf2xml tool will use, in this case 4GB, within the Grobid server, further we restrict docker to only use at most 4 CPUS run the following:

docker run --rm --init -p 127.0.0.1:8070:8070 -p 127.0.0.1:8071:8071 \
           --memory=5g --memory-swap=5g --cpus=4 \
           --env grobid__3rdparty__pdf2xml__memory__limit__mb=4000 \
           ucrel/grobid:0.6.1

Note use the --init flag when running, helps remove processes when exiting the container.⁠

An example of how to run docker container through docker-compose while restricting the amount of overall memory Docker will use, in this case 5GB, and the amount of memory the pdf2xml tool will use, in this case 4GB, within the Grobid server, further we restrict docker to only use at most 4 CPUS. This can be seen in the docker-compose file ./docker-compose.yml⁠ which can be ran with the following command:

docker-compose --compatibility up # --compatibility is required so that it restricts the overall docker memory and number of CPUs
crtl +c # Stop docker-compose after it has started
docker-compose down

⁠What is Grobid

"A machine learning library for extracting, parsing and re-structuring raw documents such as PDF into structured XML/TEI encoded documents with a particular focus on technical and scientific publications." taken from the Grobid documentation⁠

⁠Main difference between this and the original Grobid docker image

The Grobid library does have it's own Docker image (found at https://hub.docker.com/r/lfoppiano/grobid/⁠) that builds the Grobid server but it runs the Grobid server as a root user. The version within this repository runs the Grobid server as a non-root/non-privileged user and this version is slightly smaller by 0.3GB (original version 1.55GB, this version 1.22GB). The size difference is due to, in part, running as a non-root/non-privileged user.

This version also does not have a Volume mounted at (from what I gather this is where the data is temporary stored when Grobid is processing the PDF.):

/opt/grobid/grobid-home/tmp

⁠Script in detail

The script first complies the Grobid Java code using build.dockerfile⁠, then the Dockerfile⁠ uses that compiled code, by copying it accross.

If you want to use a newer version of Grobid change the GROBID_VERSION variable in the script to the release version you want from the Grobid GitHub page.⁠ Note by changing the GROBID_VERSION variable does not mean it will definitely work, it may require some changes. At the moment it currently works with Grobid version 0.6.1.

⁠Acknowledgements

The work has been funded by the UCREL research centre at Lancaster University⁠.

We would like to thank the developers of Grobid and Grobid's main author Patrice Lopez.⁠ for creating the Grobid software.⁠