How to configure backend settings of document ingestion

[djdameln]

Hi, thanks for open-sourcing this awesome project!

I have been playing around with an experimental RAG workflow, for which I want to customize the document ingestion behaviour of the DocumentSearch module. More specifically, I want to disable OCR to speed up document ingestion for faster prototyping. Effectively, this means that I need to disable OCR in the PdfPipelineOptions used by the DocumentConverter instance of the Docling backend.

However, it seems that the various sets of pipeline options used by the DocumentConverter are hardcoded in the implementation of DoclingDocumentParser. As a result, I had to create a new subclass of DocumentParser in which I customize the Docling PipelineOptions to my preferences.

Is this the correct workflow, or is there a simpler way to configure the ingestion settings to my liking? It seems a bit overkill to define a new subclass just to change a module's configuration.

(Similarly, I would like to change the chunking behaviour of my document parser, which I believe would require a similar solution).

[mhordynski]

Hi @djdameln! Sorry for the late response. I haven't properly configured notifications from GH discussions.

You are absolutely right, we should allow options to be passed to the Docling parser. I've created an issue for that, and we'll add this to the next release: #662

To answer the question:

Is this the correct workflow, or is there a simpler way to configure the ingestion settings to my liking? It seems a bit overkill to define a new subclass just to change a module's configuration.

while DocumentParser is designed to be subclassed and extended by the users - this particular case should (and will be soon) be available without it.

Thanks for contributing!