I am trying to use AutoGluon MultiModalPredictor to create a model which recognizes Facial Expressions, but using CPU is taking too long. I tried to switch to the AMD Radeon Graphics card on my PC but AutoGLuon is unable to recognize it, and as far as I can see, there is only compatibility for NVIDIA CUDA.

When downloading a GPU specific autogluon version:

mamba install -c conda-forge -c pytorch -c nvidia autogluon "pytorch=*=*cuda*"

When specifying number of GPUs in hyperparameters:

hyperparameters={'env.num_gpus':1} gives UserWarning: Using the detected GPU number 0, smaller than the GPU number 1 in the config.

Is there any way to get AutoGluon to recognize the AMD GPU?