feat (ai, model): automatically select GPU or CPU

[reycn]

Description

This PR supports automatic device selection supported by pytorch.

• If cuda is properly installed, indicated by torch.cuda.is_available(), we use GPU for inference
  • Please note that we have the same outputs of loading the model (cpu) yet in inference, GPU is utilized.
• If not, use cpu for compatibility.

Related Issue

• ckpt = torch.load(file, map_location="cpu")

Test

Test passed on RTX3080, using the following environment: conda create -n pdf python=3.11 pytorch torchvision transformers ipywidgets ipykernel -y -c nvidia -c conda-forge