Thanks a lot ! It should fix #1027 yes, we'll see for the next docker build.

I've created a new branch 0.7.3-fixes and cherry-picked these changes there.
I think we can build the 0.7.3 docker image again and replace the old one.

Both the CRF and the CRF for ARM docker images are updated.
In the HF space, in the embedded version the bibtex is still returned.. There is something related to the HF stuff that is changing the request from xml to text I think.

In the HF space, in the embedded version the bibtex is still returned.. There is something related to the HF stuff that is changing the request from xml to text I think.

With slightly older docker image, it was working, for example https://huggingface.co/spaces/kermitt2/grobid-crf running with 0.7.3.-SNAPSHOT is correctly managing XML for processHeader. I was rather suspecting something with the JDK version (more or less the only thing that changed between the images), but weird in any cases.

Sorry, good news.. it's working.. It seems that I did not properly refresh..

Cool ! I'll update the full image too.

It's done for the full image on Docker Hub too, and it works fine on HF.