Refine vllm inference and keep the API same as non-vllm #49

	name: Unit Tests

	on:
	workflow_dispatch:
	pull_request_review:
	types: [submitted]
	branches:
	- main
	paths-ignore:
	- 'docs/**'
	push:
	branches:
	- main
	paths-ignore:
	- 'docs/**'
	tags:
	- v[0-9]+.[0-9]+.[0-9]+

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

	jobs:
	run-shell-script:
	if: github.event.review.state == 'approved'
	runs-on: self-hosted

	steps:
	- name: Check for 2 approvals
	id: approvals
	run: \|
	REVIEW_API_URL="https://api.github.com/repos/${{ github.repository }}/pulls/${{ github.event.pull_request.number }}/reviews"
	reviews=$(curl -s -H "Authorization: token ${{ secrets._GITHUB_TOKEN }}" $REVIEW_API_URL)
	approve_count=$(echo "$reviews" \| jq '[.[] \| select(.state == "APPROVED")] \| length')
	echo "approve_count=$approve_count" >> $GITHUB_ENV

	- name: Checkout code
	uses: actions/checkout@v3

	- name: Run unit test
	if: env.approve_count == '2'
	run: \|
	docker pull $UT_IMAGE
	docker run -v $PWD:$PWD -w $PWD --net host --ipc host --shm-size 80G -t --rm --gpus all $UT_IMAGE bash -c 'make test'
	env:
	UT_IMAGE: ${{ secrets.UT_IMAGE }}

Provide feedback