docker-image: wrong path provided to Delft for `resources-registry.json`

[v01dXYZ]

First, thank you for the project. I found a little inconvenience with the current docker image to report.

Grobid version

docker latest-full image-id: 8a2b71b23eba

Operating System and architecture (arm64, amd64, x86, etc.)

No response

What is your Java version

No response

Log and information

load_resource_registry() seems to be called with /usr/local/lib/python3.11/dist-packages/delft/resources-registry.json instead of /opt/grobid/delft/resources-registry.json. Overriding the path in the container Embeddings.py as a quick fix solves the problem though.

Further information

No response

[lfoppiano]

@v01dXYZ which docker image are you using? I cannot find that one...

The Dockerfile looks correct to me:

RUN python3 preload_embeddings.py --registry ./resources-registry.json &&  \
    ln -s /opt/grobid /opt/delft

This docker image works, is deployed here

[v01dXYZ]

It was this one index-digest sha256:36473f3f410c0794fe25b27b4a6ed597f72f1637ed40f1c97ede36a4dbf630d5.

I confirm though problem is solved with index-digest sha256:69f9a371e9ba2e4d9f234e66884855cf437292ddb63fbed9ace9cce56d34cb81 .

Thank you very much for your fast answer.

[v01dXYZ]

@lfoppiano Actually on the fresher docker image, it is still loading the resources-registry.json from the package dir. It was not a GROBID issue but rather a DLFT issue (kermitt2/delft#201).

[lfoppiano]

This issue was fixed in delft 0.4.1 🤔 so it should be included in the latest image. Could you please share the log and the docker image that you used?

[v01dXYZ]

Yes the issue is fixed in DLFT as you put resources-registry.json in pyproject.toml. Yet it is the resources-registry.json from the package which is loaded not the one from the container. Everything runs fine, but I put in DLFT load_resources():

if not path.statswith("delft"):
    raise ValueError(f"PATH {path}")

it shows the path of the package. I don't know if it is wrong.

[lfoppiano]

Hi @v01dXYZ, yes that's OK because they are the same, for the time being (thanks for letting me know - I need to update the grobid resource-registry.json eventually).
We will fix this properly in DeLFT with kermitt2/delft#198. There is already a PR that's need to be tested also in Grobid.