modelscope-quick-start.yml

# ModelScope quick start guard - project thin trigger.
#
# Calls the common engine .github/workflows/quick-start-template.yml;
# this file only declares what varies per project: schedule,
# concurrency, runner, container image, upstream repo, monitored doc
# and the test entry command. The guard loop, cache I/O, state relay
# and publishing all live in the engine. See the engine's
# `workflow_call.inputs` block for the full input contract.
#
# Engine-fixed behaviours this trigger relies on (declared in the
# engine, NOT here):
#   * cache key prefix: monitor-state-modelscope-  (from inputs.project)
#   * artifact name:    modelscope-quick-start-<run_id>
#   * result.json is written by the engine and validated against
#     schemas/result.schema.json before upload.
#
# ModelScope note: the doc exercises the hub client itself - source
# install at the latest release tag + `modelscope download` +
# AutoModelForCausalLM text generation on NPU (modelscope's
# verify_device whitelist rejects device='npu', so pipeline() is not
# usable; see the doc's 使用样例 note).
#
# Model downloads land in /root/.cache/modelscope: no explicit bind
# mount is needed because NFS already mounts /root/.cache (same setup
# as the peft / comfyui triggers).

name: modelscope-quick-start

concurrency:
  # format() is load-bearing: a '||' between 'manual-' and
  # github.run_id would short-circuit on the truthy literal and
  # every dispatch would share one 'manual-' group.
  group: ${{ github.event_name == 'schedule' && 'modelscope-quick-start-schedule' || format('manual-{0}', github.run_id) }}
  # cancel-in-progress: false because (1) a schedule run cancelled
  # mid-way loses its outcome writeback - the outcome is what makes
  # the retry mechanism work, and the 'if: always()' guard isn't
  # enough when the container is being torn down; (2) dispatch / PR
  # runs already live in unique groups so there's nothing to cancel.
  cancel-in-progress: false

on:
  schedule:
    # Same 12 h cadence as the comfyui trigger, minute offset (55 vs 45)
    # so the two jobs don't contend for the single NPU runner slot.
    - cron: '55 */12 * * *'
  workflow_dispatch:
  # PR trigger: docs/tests changes get a guard run. paths filter avoids burning
  # the self-hosted NPU runner on unrelated PRs. `pull_request` (not
  # `pull_request_target`): contents: read is enough, no write-token risk.
  pull_request:
    branches: [main]
    paths:
      - 'sources/modelscope/**'
      - 'tests/modelscope/**'

permissions:
  contents: read

jobs:
  modelscope-quick-start:
    uses: ./.github/workflows/quick-start-template.yml
    with:
      # Namespaces cache keys (monitor-state-modelscope-*), artifacts
      # (modelscope-quick-start-<run_id>) and the test working dir
      # (tests/modelscope).
      project: modelscope
      # Self-hosted NPU runner for the test job only; the engine pins
      # the cache I/O jobs (restore-cache / publish-and-persist) to
      # GitHub-hosted ubuntu-latest.
      test_runner: '["linux-aarch64-a2-1"]'
      image: swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:9.1.0-910b-ubuntu22.04-py3.12
      # 2 h budget: cold run reuses the image torch stack + source
      # installs modelscope at the latest release tag (uv pip install -e
      # '.[framework]', whose aarch64 resolver is steered off CUDA wheels
      # by the CUDA exclusion list) + pulls Qwen2.5-0.5B-Instruct (~1 GB,
      # warm after the first run via the NFS-mounted /root/.cache) + a
      # single-card text-generation smoke on a 910B4. Covers several
      # retries on dependency flakes.
      timeout_minutes: 120
      upstream_repo: modelscope/modelscope
      # Doc URL points to the upstream Ascend/docs repo. {0} is filled by the
      # engine: PR head SHA on PR runs, 'main' otherwise. Same-repo PRs (head
      # SHA exists on Ascend/docs) test the PR-version of the doc; fork PRs
      # (head SHA on a fork) 404 on doc fetch - accepted, since the content
      # lands on Ascend/docs post-merge.
      doc_url: 'https://raw.githubusercontent.com/Ascend/docs/{0}/sources/modelscope/quick_start.md'
      doc_path: https://github.com/Ascend/docs/blob/main/sources/modelscope/quick_start.md
      # cwd is the repo root inside the checkout; env contract
      # (MONITORED_DOC_URL / UPSTREAM_REF / NPU_READY) is injected by the
      # engine. All project env prep and installs live in the test
      # subclass's prepare_environment hook.
      test_command: python -m unittest tests.modelscope.test_quick_start_ascend -v 2>&1