Lightweight installation options
Requested feature
Currently, Docling has a heavy dependency tree with dependencies such as Torch. However, not all document parsing methods need this. For example, when using Docling for docx parsing with default options, Torch (and a number of other dependencies) are not used. It would be great if there was an option for a lightweight install, with optional dependencies for certain parsing capabilities.
Alternatives
Directly using downstream libraries such as python-docx.
Thanks in advance!
We will start a docling-slim package exactly for this purpose. Stay tuned
Awesome, looking forward to it!
Also looking forward to it, as currently simple convert from html to md requires 4GB of dependencies.
The issue in which the docling-slim plan is described is:
- https://github.com/docling-project/docling/issues/2535