PR #246 - Label ci-all - id-3471466469-mhegdvgu #62
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: Examples - Unsloth | |
| permissions: | |
| contents: read | |
| on: | |
| schedule: | |
| # Every day at 5 AM UTC+8 | |
| - cron: '0 21 * * *' | |
| workflow_dispatch: | |
| repository_dispatch: | |
| types: [ci-unsloth, ci-all] | |
| run-name: >- | |
| ${{ github.event_name == 'repository_dispatch' | |
| && format( | |
| 'PR #{0} - Label {1} - {2}', | |
| github.event.client_payload.pull_number, | |
| github.event.client_payload.ci_label, | |
| github.event.client_payload.correlation_id | |
| ) | |
| || format('Unsloth - {0}', github.event_name) }} | |
| jobs: | |
| unsloth: | |
| if: > | |
| github.event_name != 'repository_dispatch' || | |
| github.event.action == 'ci-unsloth' || | |
| github.event.action == 'ci-all' | |
| name: Unsloth (Python ${{ matrix.python-version }}, ${{ matrix.setup-script }}) | |
| runs-on: [self-hosted, 1ES.Pool=agl-runner-gpu] | |
| timeout-minutes: 60 | |
| strategy: | |
| matrix: | |
| # Legacy versions are not supported for Unsloth examples. | |
| include: | |
| - python-version: '3.12' | |
| setup-script: 'stable' | |
| - python-version: '3.13' | |
| setup-script: 'latest' | |
| fail-fast: false | |
| steps: | |
| - name: Check GPU status | |
| run: nvidia-smi | |
| - name: Check disk space | |
| run: df -h | |
| - uses: actions/checkout@v4 | |
| with: | |
| ref: ${{ github.event_name == 'repository_dispatch' && github.event.client_payload.pr_ref || (github.event.pull_request.number && format('refs/pull/{0}/merge', github.event.pull_request.number)) || github.ref }} | |
| - uses: astral-sh/setup-uv@v7 | |
| with: | |
| enable-cache: true | |
| python-version: ${{ matrix.python-version }} | |
| - name: Upgrade dependencies (latest) | |
| run: uv lock --upgrade | |
| if: matrix.setup-script == 'latest' | |
| - name: Sync dependencies | |
| run: | | |
| uv sync --frozen --no-default-groups --extra verl \ | |
| --group dev --group experiment --group trl --group agents --group torch-gpu-stable | |
| - name: Freeze dependencies | |
| run: | | |
| set -ex | |
| uv pip freeze | tee requirements-freeze.txt | |
| echo "UV_LOCKED=1" >> $GITHUB_ENV | |
| echo "UV_NO_SYNC=1" >> $GITHUB_ENV | |
| - name: Upload dependencies artifact | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: dependencies-unsloth-${{ matrix.python-version }}-${{ matrix.setup-script }} | |
| path: requirements-freeze.txt | |
| compression-level: 0 | |
| - name: Prepare Unsloth model | |
| run: | | |
| set -ex | |
| cd examples/unsloth | |
| rm -rf models | |
| uv run hf download unsloth/Qwen3-4B-Instruct-2507 --local-dir models/version_0 | |
| - name: Unsloth SFT example | |
| run: | | |
| set -ex | |
| source .venv/bin/activate | |
| cd examples/unsloth | |
| agl store --port 4747 & | |
| sleep 5 | |
| python sft_rollout_runners.py & | |
| sleep 5 | |
| python sft_algorithm.py | |
| pkill -f agl && echo "SIGTERM sent to agl" || echo "No agl process found" | |
| while pgrep -f agl; do | |
| echo "Waiting for agl to finish..." | |
| sleep 5 | |
| done | |
| pkill -f sft_rollout_runners.py && echo "SIGTERM sent to sft_rollout_runners.py" || echo "No sft_rollout_runners.py process found" | |
| while pgrep -f sft_rollout_runners.py; do | |
| echo "Waiting for sft_rollout_runners.py to finish..." | |
| sleep 5 | |
| done | |
| echo "sft_rollout_runners.py has finished." | |
| sleep 10 | |
| # Check models/version_2 must exist | |
| if [ ! -d "models/version_2" ]; then | |
| echo "models/version_2 does not exist" | |
| exit 1 | |
| fi | |
| env: | |
| WANDB_BASE_URL: ${{ secrets.MSR_WANDB_BASE_URL }} | |
| WANDB_API_KEY: ${{ secrets.MSR_WANDB_API_KEY }} | |
| - name: Unsloth SFT example all-in-one | |
| run: | | |
| set -ex | |
| source .venv/bin/activate | |
| cd examples/unsloth | |
| rm -rf models/version_1 models/version_2 | |
| python sft_allinone.py | |
| if [ ! -d "models/version_2" ]; then | |
| echo "models/version_2 does not exist" | |
| exit 1 | |
| fi | |
| env: | |
| WANDB_BASE_URL: ${{ secrets.MSR_WANDB_BASE_URL }} | |
| WANDB_API_KEY: ${{ secrets.MSR_WANDB_API_KEY }} |