GPU recording for gpt-oss:20b (responses suite) #17

Workflow file for this run

.github/workflows/launch-gpu-ec2-runner.yml at afadbcc

	name: 'Launch GPU EC2 Runner'

	run-name: GPU recording for gpt-oss:20b (${{ inputs.suite }} suite)

	on:
	workflow_dispatch:
	inputs:
	suite:
	description: 'Test suite to run'
	required: false
	type: choice
	default: 'base'
	options:
	- base
	- responses
	- vllm-reasoning

	concurrency:
	group: gpu-vllm-record-${{ github.run_id }}
	cancel-in-progress: false # Don't cancel - EC2 cleanup is critical

	# OIDC authentication for AWS - no long-lived credentials!
	permissions:
	contents: read
	id-token: write # Required for OIDC authentication to AWS

	jobs:
	# Job 1: Launch GPU EC2 instance with multi-AZ fallback
	start-gpu-runner:
	runs-on: ubuntu-latest
	outputs:
	label: ${{ steps.start-ec2-runner.outputs.label }}
	instance-id: ${{ steps.start-ec2-runner.outputs.ec2-instance-id }}
	steps:
	- name: Checkout code
	uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
	with:
	persist-credentials: false

	- name: Configure AWS credentials via OIDC
	uses: aws-actions/configure-aws-credentials@e3dd6a429d7300a6a4c196c26e071d42e0343502 # v4.0.2
	with:
	role-to-assume: ${{ secrets.AWS_ROLE_ARN }}
	aws-region: us-east-2
	role-session-name: GitHubActions-vLLM-GPU-${{ github.run_id }}

	- name: Start EC2 runner
	id: start-ec2-runner
	uses: ./.github/actions/launch-gpu-runner
	with:
	mode: start
	github-token: ${{ secrets.RELEASE_PAT }}
	instance-type: g6.2xlarge
	aws-region: us-east-2
	availability-zones-config: \|
	[
	{"imageId": "${{ vars.AWS_EC2_AMI_US_EAST_2 }}", "subnetId": "${{ vars.SUBNET_US_EAST_2A }}", "securityGroupId": "${{ vars.SECURITY_GROUP_ID_US_EAST_2 }}"},
	{"imageId": "${{ vars.AWS_EC2_AMI_US_EAST_2 }}", "subnetId": "${{ vars.SUBNET_US_EAST_2B }}", "securityGroupId": "${{ vars.SECURITY_GROUP_ID_US_EAST_2 }}"},
	{"imageId": "${{ vars.AWS_EC2_AMI_US_EAST_2 }}", "subnetId": "${{ vars.SUBNET_US_EAST_2C }}", "securityGroupId": "${{ vars.SECURITY_GROUP_ID_US_EAST_2 }}"}
	]
	ec2-instance-tags: \|
	[
	{"Key": "Name", "Value": "llamastack-vllm-gpu-runner"},
	{"Key": "Project", "Value": "llama-stack"},
	{"Key": "Purpose", "Value": "vllm-gpu-recording"},
	{"Key": "Model", "Value": "gpt-oss:20b"},
	{"Key": "GitHubRepository", "Value": "${{ github.repository }}"},
	{"Key": "GitHubRef", "Value": "${{ github.ref }}"},
	{"Key": "GitHubRunId", "Value": "${{ github.run_id }}"},
	{"Key": "ManagedBy", "Value": "GitHub-Actions"}
	]

	- name: Runner launch summary
	run: \|
	echo "GPU runner launched successfully"
	echo " Instance ID: ${{ steps.start-ec2-runner.outputs.ec2-instance-id }}"
	echo " Runner Label: ${{ steps.start-ec2-runner.outputs.label }}"
	echo " Model: gpt-oss:20b"
	echo " Instance Type: g6.2xlarge"

	# Job 2: Run vLLM tests on GPU runner
	record-vllm-tests:
	needs: start-gpu-runner
	runs-on: ${{ needs.start-gpu-runner.outputs.label }}
	permissions: {} # CRITICAL: No permissions - prevents secret theft from untrusted code
	env:
	TMPDIR: /home/tmp
	steps:
	- name: Checkout code
	uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
	with:
	persist-credentials: false

	- name: Setup environment
	run: \|
	mkdir -p /home/tmp
	echo "=== System Information ==="
	cat /etc/os-release
	echo ""
	echo "=== Disk Space ==="
	df -h
	echo ""
	echo "=== Memory ==="
	free -h
	echo ""
	echo "=== GPU Information ==="
	nvidia-smi

	- name: Setup vLLM GPU
	uses: ./.github/actions/setup-vllm-gpu
	with:
	model: 'gpt-oss:20b'
	port: '8000'
	gpu-memory-utilization: '0.85'
	max-model-len: '8192'
	quantization: 'none'
	vllm-version: '0.22.1'

	- name: Setup test environment
	uses: ./.github/actions/setup-test-environment
	with:
	python-version: '3.12'
	client-version: 'latest'
	setup: 'vllm-gpu-gpt-oss'
	suite: ${{ inputs.suite }}
	inference-mode: 'record'
	enable-hf-cache: 'false'

	- name: Run integration tests (record mode)
	uses: ./.github/actions/run-and-record-tests
	with:
	stack-config: 'server:ci-tests'
	setup: 'vllm-gpu-gpt-oss'
	inference-mode: 'record'
	suite: ${{ inputs.suite }}
	pattern: ${{ inputs.suite == 'base' && 'not (test_openai_completion_guided_choice or test_multiple_tools_with_different_schemas or test_mcp_invocation)' \|\| '' }}
	skip-commit: 'true' # Don't commit here - upload as artifacts

	- name: Upload recordings as artifacts
	if: always()
	uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
	with:
	name: vllm-gpu-recordings-${{ github.run_id }}
	path: \|
	tests/integration/recordings/
	tests/integration/*/recordings/
	retention-days: 7
	if-no-files-found: error

	- name: Upload vLLM logs
	if: always()
	run: \|
	if [ -f /tmp/vllm-server.log ]; then
	cat /tmp/vllm-server.log
	fi

	- name: Disk space after tests
	if: always()
	run: \|
	echo "=== Disk Space After Tests ==="
	df -h

	# Job 3: Stop GPU EC2 instance (ALWAYS runs for cleanup)
	stop-gpu-runner:
	needs: [start-gpu-runner, record-vllm-tests]
	runs-on: ubuntu-latest
	if: ${{ always() }} # CRITICAL: Always cleanup, even on failure or cancellation
	steps:
	- name: Checkout code
	uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
	with:
	persist-credentials: false

	- name: Configure AWS credentials via OIDC
	uses: aws-actions/configure-aws-credentials@e3dd6a429d7300a6a4c196c26e071d42e0343502 # v4.0.2
	with:
	role-to-assume: ${{ secrets.AWS_ROLE_ARN }}
	aws-region: us-east-2
	role-session-name: GitHubActions-vLLM-GPU-Cleanup-${{ github.run_id }}

	- name: Stop EC2 runner
	uses: ./.github/actions/launch-gpu-runner
	with:
	mode: stop
	github-token: ${{ secrets.RELEASE_PAT }}
	aws-region: us-east-2
	label: ${{ needs.start-gpu-runner.outputs.label }}
	ec2-instance-id: ${{ needs.start-gpu-runner.outputs.instance-id }}

	- name: Cleanup summary
	run: \|
	echo "GPU runner terminated successfully"
	echo " Instance ID: ${{ needs.start-gpu-runner.outputs.instance-id }}"

	# Job 4: Summary and next steps
	summary:
	needs: [start-gpu-runner, record-vllm-tests, stop-gpu-runner]
	runs-on: ubuntu-latest
	if: always()
	steps:
	- name: Workflow summary
	run: \|
	{
	echo "## vLLM GPU Recording Summary"
	echo ""
	echo "Model: gpt-oss:20b"
	echo "Instance Type: g6.2xlarge"
	echo "Test Suite: ${{ inputs.suite }}"
	echo ""

	if [ "${{ needs.record-vllm-tests.result }}" == "success" ]; then
	echo "Test Status: Successful"
	echo ""
	echo "Recordings have been uploaded as artifacts. Download them from the workflow run and commit manually."
	else
	echo "Test Status: Failed"
	echo ""
	echo "Check the test logs for errors."
	fi

	echo ""
	echo "Cleanup Status: ${{ needs.stop-gpu-runner.result == 'success' && 'Instance terminated' \|\| 'Check manually' }}"
	} >> "$GITHUB_STEP_SUMMARY"

	- name: Check for cleanup issues
	if: needs.stop-gpu-runner.result != 'success'
	run: \|
	echo "::warning::EC2 instance cleanup may have failed! Check AWS console for orphaned instances."
	echo "Instance ID: ${{ needs.start-gpu-runner.outputs.instance-id }}"

Navigation Menu

Uh oh!

GPU recording for gpt-oss:20b (responses suite) #17

Workflow file

GPU recording for gpt-oss:20b (responses suite) #17

Uh oh!

Workflow file for this run