Add user release benchmark so that we can run it on pull request #5

	name: Release TorchBench Userbenchmark on A100
	on:
	pull_request:
	paths:
	- userbenchmark/release-test/*

	jobs:
	run-userbenchmark:
	runs-on: [a100-runner]
	timeout-minutes: 1440 # 24 hours
	environment: docker-s3-upload
	env:
	BASE_CONDA_ENV: "torchbench"
	CONDA_ENV: "userbenchmark-a100"
	PLATFORM_NAME: "gcp_a100"
	SETUP_SCRIPT: "/workspace/setup_instance.sh"
	steps:
	- name: Checkout TorchBench
	uses: actions/checkout@v3
	with:
	path: benchmark
	ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha \|\| github.sha }}
	- name: Tune Nvidia GPU
	run: \|
	sudo nvidia-smi -pm 1
	sudo nvidia-smi -ac 1215,1410
	nvidia-smi
	- name: Clone and setup conda env
	run: \|
	CONDA_ENV=${BASE_CONDA_ENV} . "${SETUP_SCRIPT}"
	conda create --name "${CONDA_ENV}" --clone "${BASE_CONDA_ENV}"
	- name: Install TorchBench
	run: \|
	set -x
	. "${SETUP_SCRIPT}"
	pushd benchmark
	python install.py
	- name: Run user benchmark
	run: \|
	set -x
	. "${SETUP_SCRIPT}"
	# remove old results
	if [ -d benchmark-output ]; then rm -Rf benchmark-output; fi
	pushd benchmark
	release_version=$(cat userbenchmark/release-test/version.txt)
	if [ -d .userbenchmark ]; then rm -Rf .userbenchmark; fi
	python run_benchmark.py release-test -c ${release_version}
	cp -r ./.userbenchmark/release-test ../benchmark-output

	- name: Upload artifact
	uses: actions/upload-artifact@v3
	with:
	name: TorchBench result
	path: benchmark-output/
	- name: Clean up Conda env
	if: always()
	run: \|
	. "${SETUP_SCRIPT}"
	conda deactivate && conda deactivate
	conda remove -n "${CONDA_ENV}" --all

Provide feedback