Add support for `amd-smi` to perform device isolation #49

	name: OnnxRuntime TensorRT Inference Tests

	on:
	pull_request:
	types: [opened, reopened, synchronize]

	concurrency:
	group: ${{ github.workflow }}-${{ github.event.pull_request.number \|\| github.ref }}
	cancel-in-progress: true

	jobs:
	build_image_and_run_gpu_tests:
	runs-on: hf-dgx-01
	steps:
	- name: Checkout
	uses: actions/checkout@v2

	- name: Build image
	run: docker build
	--file docker/tensorrt.dockerfile
	--build-arg USER_ID=$(id -u)
	--build-arg GROUP_ID=$(id -g)
	--build-arg TENSORRT_VERSION=22.12
	--build-arg TORCH_CUDA=cu118
	--tag opt-bench-tensorrt:22.12
	.

	- name: Run tests
	run: docker run
	--rm
	--gpus '"device=0,1"'
	--entrypoint /bin/bash
	--volume $(pwd):/workspace/optimum-benchmark
	--workdir /workspace/optimum-benchmark
	--env USE_CUDA="1"
	opt-bench-tensorrt:22.12
	-c "pip install -e .[test,onnxruntime-gpu,diffusers] && pytest -k 'tensorrt and onnxruntime and inference' -x"

Provide feedback