still cannot achieve GPU inference on linux or mac

[sammlapp]

Describe the bug I have an environment on a linux machine in which other tensorflow models run on my GPU with cuda 12.2, torch 2.15, and cudnn 8.9. With other TF models, as soon as I load the model, memory on the GPU is used by Tensorflow. With BirdNET, no memory is used and the process only runs on my CPU.

If anyone can successfully use the models provided in this repo on GPU, can you provide an example of your package versions and code?

Expected behavior GPU used when model is loaded