Which model are you using for the demo?
because I use ibm-granite/granite-docling-258M model to extract data but It can't work as well , but in demo it's ok for extract formula and data of table.
Hi
@MelikaNLP
, which demo are you referring to? There is the WebGPU demo which uses just the model, and the full Docling demo which uses the full docling library. If you're comparing to the later, it's likely that you'll get different results based on the additional processing on both input and output. If you're comparing to the WebGPU demo, it's likely that there are some subtle differences in your pre/post processing causing the differences. You can check out the code details for the WebGPU demo here: https://huggingface.co/spaces/ibm-granite/granite-docling-258M-WebGPU/blob/main/index.html#L203