Make the CI stable (no random failures), faster (8 intel cpus) and better orchestrated #1294

	name: GPTQ Quantization / Test GPU

	on:
	workflow_dispatch:
	schedule:
	- cron: 0 1 /3 * # at 1am every 3 days
	pull_request:
	types: [opened, synchronize, reopened, labeled]
	# uncomment to enable on PR merge on main branch:
	#push:
	# branches:
	# - main

	jobs:
	do-the-job:
	if: ${{ (github.event_name == 'workflow_dispatch') \|\| (github.event_name == 'schedule') \|\| contains( github.event.pull_request.labels.*.name, 'gpu-test') }}
	name: Start self-hosted EC2 runner
	runs-on: [single-gpu, nvidia-gpu, t4, ci]
	env:
	AWS_REGION: us-east-1
	steps:
	- name: Checkout
	uses: actions/checkout@v2
	- name: Build image
	run: \|
	docker build -f tests/gptq/Dockerfile_quantization_gpu -t gptq-gpu .
	- name: Test with unittest within docker container
	run: \|
	docker run --rm --gpus all -v $(pwd)/hf_cache:/root/.cache/huggingface --workdir=/workspace/optimum/tests gptq-gpu:latest

Provide feedback