Issues Running custom_convert Example with Local Models (Python 3.11, Docling 2.23, Ubuntu 22)

[mugeshk97]

Hello,

I’m trying to run the custom_convert notebook example from the Docling documentation (link) after downloading the models locally. However, I am encountering no progress in the execution.

Here are the details of my setup:

Python version: 3.11
Docling version: 2.23
Operating system: Ubuntu 22.04

Despite following the steps in the notebook and ensuring that all dependencies are correctly installed, the process does not seem to move forward. Could anyone guide me on what might be causing this or suggest any troubleshooting steps?

Thanks in advance

[dolfim-ibm]

I think the process started and will terminate. How many pages do you have in your document?

[mugeshk97]

Just 3 pages of 531 KB

[dolfim-ibm]

This is weird. Do you need something specific from the python example code?

Otherwise you could try running the CLI. You could also use the -v or -vv flags for more verbose output.

[mugeshk97]

I'm experiencing issues with running the GPU version of Docling. The CPU version works fine, but when I try to use the GPU, the process does not progress.

System Details:
GPU: NVIDIA GeForce GTX 1650
CUDA: 11.8
cuDNN: 8.9.7
Python: 3.11
Docling: 2.23
OS: Ubuntu 22.04
I have verified that my CUDA and cuDNN installations are correct, and other GPU-based frameworks (like PyTorch) are working as expected. However, Docling does not seem to utilize the GPU.

Is there any additional setup required to enable GPU support for Docling? Any insights or troubleshooting steps would be greatly appreciated.

Thanks!

[dolfim-ibm]

I see you are using CUDA 11.8 which is not anymore the default for the pypi distribution.

Can you please ensure you installed torch and torchvision via the documented steps described on https://pytorch.org/?

[mugeshk97]

Thanks for the input issue is resolved by specific torch distribution