Nightly GPU Integration #5
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Nightly GPU Integration | |
| on: | |
| schedule: | |
| # 01:00 PST / 02:00 PDT. GitHub schedules always use UTC. | |
| - cron: "0 9 * * *" | |
| workflow_dispatch: | |
| permissions: | |
| contents: read | |
| concurrency: | |
| group: nightly-gpu-integration | |
| cancel-in-progress: false | |
| jobs: | |
| changes: | |
| name: Check recent code changes | |
| runs-on: ubuntu-latest | |
| outputs: | |
| should_run: ${{ steps.detect.outputs.should_run }} | |
| steps: | |
| - name: Checkout main history | |
| uses: actions/checkout@fbc6f3992d24b796d5a048ff273f7fcc4a7b6c09 # v5 | |
| with: | |
| fetch-depth: 0 | |
| persist-credentials: false | |
| - name: Detect relevant changes in the preceding 24 hours | |
| id: detect | |
| shell: bash | |
| run: | | |
| set -euo pipefail | |
| if [[ "${{ github.event_name }}" == workflow_dispatch ]]; then | |
| echo "should_run=true" >> "$GITHUB_OUTPUT" | |
| echo "Manual dispatch forces the nightly integration." >> "$GITHUB_STEP_SUMMARY" | |
| exit 0 | |
| fi | |
| mapfile -t changed_files < <( | |
| git log --since="24 hours ago" --name-only --pretty=format: -- \ | |
| torchspec configs tools tests docker pyproject.toml uv.lock \ | |
| .github/workflows/gpu-integration.yml \ | |
| .github/workflows/gpu-nightly.yml \ | |
| | sed "/^[[:space:]]*$/d" | sort -u | |
| ) | |
| if ((${#changed_files[@]} == 0)); then | |
| echo "should_run=false" >> "$GITHUB_OUTPUT" | |
| echo "No relevant code changed in the preceding 24 hours; GPU integration skipped." \ | |
| >> "$GITHUB_STEP_SUMMARY" | |
| exit 0 | |
| fi | |
| echo "should_run=true" >> "$GITHUB_OUTPUT" | |
| { | |
| echo "Relevant changes found in the preceding 24 hours:" | |
| printf -- '- %s\n' "${changed_files[@]}" | |
| } >> "$GITHUB_STEP_SUMMARY" | |
| integration: | |
| name: Qwen3.8 full-epoch GPU training | |
| needs: changes | |
| if: needs.changes.outputs.should_run == 'true' | |
| runs-on: ${{ vars.TORCHSPEC_GPU_DISPATCHER_LABEL }} | |
| timeout-minutes: 720 | |
| env: | |
| TORCHSPEC_CI_MODEL: Qwen/Qwen3.8-27B | |
| TORCHSPEC_CI_MODEL_REVISION: 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0 | |
| TORCHSPEC_CI_SOURCE_SHA: ${{ github.sha }} | |
| TORCHSPEC_SLURM_ROOT: ${{ secrets.TORCHSPEC_SLURM_ROOT }} | |
| TORCHSPEC_CI_IMAGE: ${{ secrets.TORCHSPEC_CI_IMAGE }} | |
| TORCHSPEC_CI_MODEL_CACHE_HOST: ${{ secrets.TORCHSPEC_CI_MODEL_CACHE_HOST }} | |
| steps: | |
| - name: Checkout trusted dispatcher | |
| uses: actions/checkout@fbc6f3992d24b796d5a048ff273f7fcc4a7b6c09 # v5 | |
| with: | |
| ref: ${{ github.workflow_sha }} | |
| path: trusted | |
| persist-credentials: false | |
| - name: Checkout nightly source without credentials | |
| uses: actions/checkout@fbc6f3992d24b796d5a048ff273f7fcc4a7b6c09 # v5 | |
| with: | |
| ref: ${{ github.sha }} | |
| path: source | |
| persist-credentials: false | |
| - name: Submit and wait for Slurm integration | |
| run: >- | |
| trusted/tools/ci/dispatch_slurm.sh | |
| "$GITHUB_WORKSPACE/source" | |
| "$RUNNER_TEMP/slurm-report" | |
| - name: Publish Slurm summary | |
| if: always() | |
| run: | | |
| if [[ -f "$RUNNER_TEMP/slurm-report/summary.md" ]]; then | |
| cat "$RUNNER_TEMP/slurm-report/summary.md" >> "$GITHUB_STEP_SUMMARY" | |
| else | |
| echo "Slurm dispatcher exited before producing a report." >> "$GITHUB_STEP_SUMMARY" | |
| fi | |
| - name: Upload nightly artifacts | |
| if: always() | |
| uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4 | |
| with: | |
| name: nightly-gpu-${{ github.sha }}-${{ github.run_attempt }} | |
| path: ${{ runner.temp }}/slurm-report | |
| if-no-files-found: warn |