docling-serve: Why smovlm and not smoldocling?

[thistleknot]

isn't smodocling made for docling?

INFO: 127.0.0.1:51899 - "GET /v1alpha/status/poll/32e34cdd-7385-4bf4-bb57-2b3bec316137?wait=5 HTTP/1.1" 200 OK
ERROR:docling_serve.gradio_ui:Error processing file(s): Task failed with status 'failure'
Error while downloading from https://huggingface.co/HuggingFaceTB/SmolVLM-256M-Instruct/resolve/7e3e67edbbed1bf9888184d9df282b700a323964/model.safetensors: [SSL: DECRYPTION_FAILED_OR_BAD_RECORD_MAC] decryption failed or bad record mac (_ssl.c:2536)

I think I can solve the error, but curious why SmolVLM-256M vs SmolDocling?

[dolfim-ibm]

SmolDocling is actually a fine-tuning of SmolVLM, which introduces the concept of DocTags for converting documents. You can read more in the respective paper.

I think you are referring to the picture description feature in Docling. This functionality doesn't convert the document (i.e. the SmolDocling task) but it is only making a textual description of the figures. For this task you can leverage almost any vision-language model (SmolVLM, Granite Vision, etc)