How to enable GPU usage for docling?

[Stdev95]

Question

Hi,

I'm exploring docling and tried doing some experiments but I see that it takes a lot of time to parse and convert the doc to markdown or json format.
I'm trying to convert pdf to markdown with these config:
pipeline_options = PdfPipelineOptions() pipeline_options.do_ocr = True pipeline_options.do_table_structure = True pipeline_options.ocr_options.use_gpu = True pipeline_options.table_structure_options.do_cell_matching = True

I have enabled the option to use GPU but it seems that GPU is not being used.

my observation:
it is taking 650+ sec to process a pdf file of size 18MB (400+ pages) using above configurations for the pipeline.

I have two questions here:

1. for EasyOCR, do we need to enable the flag for GPU usage or default is True?
2. Do we need to follow any steps to use local GPU for processing?

[cau-git]

@Stdev95 By default, Docling already uses a GPU when detected on most models (TableFormer, EasyOCR). You don't need to enable anything specific. We have work-in-progress on complete GPU support for all models, including the layout model (see DS4SD/docling-ibm-models#50) and will inform on our README and Changelog when it is available.