Add the logic for Energy Star #527

	name: CLI ROCm Pytorch Multi-GPU Tests

	on:
	workflow_dispatch:
	push:
	branches:
	- main
	paths:
	- .github/workflows/test_cli_rocm_pytorch_multi_gpu.yaml
	- "optimum_benchmark/**"
	- "docker/**"
	- "tests/**"
	- "setup.py"
	pull_request:
	branches:
	- main
	paths:
	- .github/workflows/test_cli_rocm_pytorch_multi_gpu.yaml
	- "optimum_benchmark/**"
	- "docker/**"
	- "tests/**"
	- "setup.py"

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

	jobs:
	run_cli_rocm_pytorch_multi_gpu_tests:
	runs-on: [self-hosted, amd-gpu, multi-gpu]

	container:
	image: ghcr.io/huggingface/optimum-benchmark:latest-rocm
	options: --ipc host
	--shm-size "16gb"
	--group-add video
	--device /dev/kfd
	--device /dev/dri
	--env ROCR_VISIBLE_DEVICES
	--env HIP_VISIBLE_DEVICES=0,1
	--volume /mnt/cache/.cache/huggingface:/mnt/cache/

	steps:
	- name: Checkout code
	uses: actions/checkout@v4

	- name: Install dependencies
	run: \|
	pip install -e .[testing,diffusers,timm,deepspeed,peft,autoawq,auto-gptq] "deepspeed<0.15"

	- name: Run tests
	run: \|
	pytest -x -s -k "cli and cuda and pytorch and (dp or ddp or device_map or deepspeed) and not bnb"

Provide feedback