Add Gaudi hpu accelerator option to BaseCAM

[daniel-de-leon-user293]

Hello!

This PR adds the option for users to use BaseCAM with Intel's Gaudi platform. The htcore import and mark_step() call are all that is needed in base_cam.py and I've added the 'hpu' arg in cam.py.

It should be noted that Gaudi needs a few (depending on model and input feature size) warmup executions before seeing the best computation time and is best for running GradCAM on large batches.

Steps to use cam.py on a Gaudi platform

Option 1. If you do not have access to a Gaudi locally

a. Use the free Tiber Cloud
b. In a terminal tab on the Jupyter lab interface, clone this branch

git clone -b daniel/enable_gaudi https://github.com/daniel-de-leon-user293/pytorch-grad-cam.git

c. Install dependencies

cd pytorch-grad-cam
pip install -e .

d. Run cam.py

python3 cam.py --device hpu

Option 2. If you do have access to a Gaudi locally

a. Clone this branch on your Gaudi machine

git clone -b daniel/enable_gaudi https://github.com/daniel-de-leon-user293/pytorch-grad-cam.git

b. Start a Gaudi container that mounts this branch

docker run -d --rm -it --name gradcam-hpu -v pytorch-grad-cam:/workdir --runtime=habana \ 
-e http_proxy=$http_proxy -e https_proxy=$https_proxy \ 
-e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none \ 
--cap-add=sys_nice --net=host --ipc=host \ 
vault.habana.ai/gaudi-docker/1.18.0/ubuntu22.04/habanalabs/pytorch-installer-2.2.2:latest

c. Exec into the interactive container

docker exec -it gradcam-hpu bash

d. Install dependencies

cd pytorch-grad-cam
pip install -e .

e. Run cam.py

python3 cam.py --device hpu

Thank you and look forward to hearing your feedback!