Please add some description for potential GPU support.

[SvenSchoene]

trafficstars

We wonder if Unstructured is benefitting from deployment on machines with GPUs?

I see some hints that Unstructured may have some GPU support:

• https://github.com/search?q=repo%3AUnstructured-IO%2Funstructured%20gpu&type=code
• Maybe it's only for PaddlePaddle that is under the Unstructured-IO umbrella? (https://github.com/Unstructured-IO/unstructured.Paddle)
• "GPU" is mentioned once in this document, but probably not directly related to using Unstructured: https://unstructured.io/blog/speeding-up-vision-transformers

Can you make it clear somehow, whether GPU is supported / if there are any benefits of deploying Unstructured on a machine with a GPU? And if so: What are those benefits? And how do we enable them explicitly?

And, on the other hand: Maybe it would make sense to clearly state somewhere that an available GPU currently makes no difference when using Unstructured.

[SvenSchoene]

Some findings:

• It seems that Table Transformers are benefitting from GPU support: https://github.com/Unstructured-IO/unstructured-inference/blob/main/unstructured_inference/models/tables.py#L52
• Donut: https://github.com/Unstructured-IO/unstructured-inference/blob/main/unstructured_inference/models/donut.py#L30
• And Chipper (which is only available for paid API users, I believe): https://github.com/Unstructured-IO/unstructured-inference/blob/main/unstructured_inference/models/chipper.py#L83

[maplessssy]

+1

[muazhari]

+9999

[netapy]

+1