Offline Reference of Model "HuggingFaceTB--SmolVLM-256M-Instruct" from local artifacts path is not working.

[gaurav-kumar-5]

When using below code by specifying local artifacts path, the following error encountered -

pipeline_options.do_picture_description=True
pipeline_options.picture_description_options = smolvlm_picture_description
pipeline_options.picture_description_options.prompt = (
        "Describe the image in three sentences. Be consise and accurate."
    )

Couldn't connect to 'https://huggingface.co' to load this file, couldn't find it in the cached files and it looks like HuggingFaceTB/SmolVLM-256M-Instruct is not the path to a directory containing a file named config.json.
Checkout your internet connection or see how to run the library in offline mode at 'https://huggingface.co/docs/transformers/installation#offline-mode'.

While below models works from local artifacts path -

1. ds4sd--docling-models
2. ds4sd--CodeFormula
3. EasyOcr
4. ds4sd--DocumentFigureClassifier

[dolfim-ibm]

Can you please check which version of Docling you are using?
The latest version is downloading the artifacts correctly.

We also have an updated documentation page for it: https://ds4sd.github.io/docling/usage/#model-prefetching-and-offline-usage

[gaurav-kumar-5]

Yes I m using the updated path as of docling version == 2.20.0

[dolfim-ibm]

How did you download the models? Using the new utility docling-tools models download?

[gaurav-kumar-5]

Nope...I did using this and got 5 models downloaded in new artifacts path

artifacts_path = StandardPdfPipeline.download_models_hf()

[dolfim-ibm]

That method should have triggered a deprecation warning. We suggest the new method moving forward.

Note that we also have a function doing the same as the SDK, see from docling.utils.model_downloader import download_models.