Support for ds4sd/docling-models in WebLLM

[iamthehimansh]

Hello WebLLM team,

I’m currently working on porting Docling to JavaScript for fully in-browser document processing — no server inference.
As part of this, I’m exploring whether the ds4sd/docling-models family can run with WebLLM.

I’d like to know if:

1. WebLLM currently supports these models directly, or
2. There are any known compatibility issues (model architecture, tokenizer format, quantization requirements).

If not supported yet, could you provide guidance on:

• Whether it’s possible to convert these models into a WebLLM-compatible format.
• Any tools, scripts, or examples for integrating custom Hugging Face models into WebLLM.

Goal: Bring Docling’s document understanding capabilities to the browser using WebLLM, enabling offline, client-side document parsing.

Thanks for your time and for building such an awesome project!