Where are the models stored by default?

[jason-riddle]

I'm looking at these two lines in the project's Dockerfile

https://github.com/DS4SD/docling/blob/main/Dockerfile#L19

RUN python -c 'from deepsearch_glm.utils.load_pretrained_models import load_pretrained_nlp_models; load_pretrained_nlp_models(verbose=True);'

RUN python -c 'from docling.pipeline.standard_pdf_pipeline import StandardPdfPipeline; StandardPdfPipeline.download_models_hf(force=True);'

When these 2 command are ran, where are the respective models being downloaded to? Are they being saved to $HF_HOME or somewhere else?

[dolfim-ibm]

The first line is currently not downloading anything. We should actually remove it from the example Dockerfile.

The second line is downloading the model weights with the Hugging Face Hub library. There are many options, see the documentation.

In this case, they should be in ~/.cache/huggingface/hub/.

Additionally, that

1. The function StandardPdfPipeline.download_models_hf() actually returns the location. So a code like this would actually print it out.

   artifacts_path = StandardPdfPipeline.download_models_hf()
   print(artifacts_path)

2. StandardPdfPipeline.download_models_hf(artifacts_path="MY PATH") accept the target location. If you specify this input argument, then you have to set it everytime you initialize the StandardPdfPipeline. See more details in the docs.