[CI] enable cpu test parallel with pre-commit

.github/workflows/_selected_tests.yaml

@@ -105,7 +105,6 @@ jobs:
         UV_SYSTEM_PYTHON: 1
     steps:
       - name: Check NPU availability
-        if: ${{ matrix.group.npu_type != 'cpu' }}
         run: |
           npu-smi info
 
@@ -160,7 +159,6 @@ jobs:
 
       - name: Get csrc hash
         id: get_csrc_hash
-        if: ${{ matrix.group.npu_type != 'cpu' }}
         run: |
           # Fail closed so a git error cannot be converted into sha256(empty)
           # and then appear to be a legitimate cache miss.
@@ -180,7 +178,6 @@ jobs:
 
       - name: Get architecture
         id: get_arch
-        if: ${{ matrix.group.npu_type != 'cpu' }}
         run: |
           ARCH=$(uname -m)
           case "$ARCH" in
@@ -191,7 +188,6 @@ jobs:
 
       - name: Restore vllm-ascend csrc cache
         id: cache-csrc
-        if: ${{ matrix.group.npu_type != 'cpu' }}
         uses: runs-on/cache/restore@v5
         with:
           path: |
@@ -206,7 +202,6 @@ jobs:
           fail-on-cache-miss: false
 
       - name: Install Mooncake wheel
-        if: ${{ matrix.group.npu_type != 'cpu' }}
         run: |
           set -euxo pipefail
 
@@ -224,7 +219,6 @@ jobs:
           pip show mooncake-transfer-engine-npu || true
 
       - name: Install vllm-project/vllm-ascend with device
-        if: ${{ matrix.group.npu_type != 'cpu' }}
         env:
           CSRC_CACHE_HIT: ${{ steps.cache-csrc.outputs.cache-hit }}
         run: |
@@ -258,23 +252,12 @@ jobs:
                 [n for n in ("chunk_gated_delta_rule_fwd_h", "chunk_fwd_o") if hasattr(ops, n)])
           PY
 
-      - name: Install vllm-project/vllm-ascend no device
-        if: ${{ matrix.group.npu_type == 'cpu' }}
-        env:
-          SOC_VERSION: ascend910b1
-          COMPILE_CUSTOM_KERNELS: 0
-        run: |
-          pip install uc-manager
-          uv pip install -r requirements-dev.txt
-          uv pip install -e .
-
       - name: Uninstall Triton for 310P tests
-        if: ${{ matrix.group.npu_type != 'cpu' && matrix.group.npu_type == '310p' }}
+        if: ${{ matrix.group.npu_type == '310p' }}
         run: |
           pip uninstall -y triton-ascend triton
 
       - name: Run selected tests with device
-        if: ${{ matrix.group.npu_type != 'cpu' }}
         continue-on-error: ${{ inputs.continue_on_error }}
         env:
           VLLM_WORKER_MULTIPROC_METHOD: spawn
@@ -295,25 +278,8 @@ jobs:
             ${TIMING_FLAG} \
             ${{ matrix.group.tests }}
 
-      - name: Run selected tests without device
-        if: ${{ matrix.group.npu_type == 'cpu' }}
-        continue-on-error: ${{ inputs.continue_on_error }}
-        env:
-          VLLM_WORKER_MULTIPROC_METHOD: spawn
-          TORCH_DEVICE_BACKEND_AUTOLOAD: 0
-          ENABLE_COVERAGE: ${{ inputs.enable-coverage}}
-        run: |
-          if [ "${{ inputs.enable-coverage }}" = "true" ]; then
-            export ENABLE_COVERAGE=true
-          fi
-          .github/workflows/scripts/run_selected_tests.sh \
-            "${{ matrix.group.npu_type }}" \
-            "${{ matrix.group.num_npus }}" \
-            "without-device" \
-            ${{ matrix.group.tests }}
-
       - name: Upload timing data
-        if: ${{ inputs.upload_timing && matrix.group.npu_type != 'cpu' }}
+        if: ${{ inputs.upload_timing }}
         continue-on-error: true
         uses: actions/upload-artifact@v7
         with:

.github/workflows/_selected_tests_cpu.yaml

@@ -0,0 +1,101 @@
+name: Selected CPU Tests
+
+on:
+  workflow_call:
+    inputs:
+      vllm:
+        description: vllm version string
+        required: true
+        type: string
+      ref:
+        description: PR head sha
+        required: true
+        type: string
+      base_ref:
+        description: target base branch name e.g. main
+        required: true
+        type: string
+      # -------- fix start --------
+      base_sha:
+        description: base commit sha, optional. empty skip rebase
+        required: false
+        default: ""
+        type: string
+      # -------- fix end --------
+      test_groups:
+        description: json test groups
+        required: true
+        type: string
+    outputs:
+      coverage_artifact_name:
+        value: ${{ jobs.run_tests.outputs.coverage_artifact_name }}
+
+jobs:
+  run_tests:
+    name: cpu‑ut‑run
+    runs-on: ${{ fromJson(inputs.test_groups)[0].runner_label }}
+    outputs:
+      coverage_artifact_name: cpu‑coverage‑report
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref }}
+          fetch-depth: 0
+
+      - name: Install python dependencies
+        run: |
+          set -euo pipefail
+          python3 -m ensurepip --upgrade
+          python3 -m pip install --upgrade pip
+          python3 -m pip install pytest pytest-cov
+
+      - name: Resolve source snapshot (optional rebase)
+        run: |
+          set -euo pipefail
+          BASE_SHA="${{ inputs.base_sha }}"
+          if [ -n "${BASE_SHA}" ]; then
+            git config user.name "ci‑runner"
+            git config user.email "ci@runner.local"
+            git rebase "${BASE_SHA}"
+          else
+            echo "INFO: base_sha is empty, skip rebase, use HEAD directly"
+          fi
+
+      - name: Parse test groups
+        id: parse_groups
+        run: |
+          set -euo pipefail
+          python3 - <<'PY'
+          import json
+          import os
+          groups = json.loads(os.environ["TEST_GROUPS"])
+          paths = []
+          for g in groups:
+              paths.extend(g["test_paths"])
+          out_paths = " ".join(paths)
+          with open(os.environ["GITHUB_OUTPUT"], "a", encoding="utf‑8") as f:
+              f.write(f"test_paths={out_paths}\n")
+          PY
+        env:
+          TEST_GROUPS: ${{ inputs.test_groups }}
+
+      - name: Setup python & dependencies
+        run: |
+          set -euo pipefail
+          python3 -m pip install --upgrade pip
+          python3 -m pip install pytest pytest‑cov
+
+      - name: Run pytest with coverage
+        run: |
+          set -euo pipefail
+          TEST_PATHS="${{ steps.parse_groups.outputs.test_paths }}"
+          echo "Running tests: ${TEST_PATHS}"
+          python3 -m pytest -v --cov=./ --cov‑report=xml:coverage.xml ${TEST_PATHS}
+
+      - name: Upload coverage artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: cpu‑coverage‑report
+          path: coverage.xml
+          retention-days: 7

.github/workflows/pr_test.yaml

@@ -179,6 +179,67 @@ jobs:
             echo "============================"
           done
 
+  run-selected-tests-cpu:
+    needs: pre-commit
+    strategy:
+      fail-fast: false
+      matrix:
+        vllm_version:
+          - ${{ needs.pre-commit.outputs.main_commit }}
+    uses: ./.github/workflows/_selected_tests_cpu.yaml
+    with:
+      vllm: ${{ matrix.vllm_version }}
+      ref: ${{ github.event.pull_request.head.sha }}
+      base_ref: ${{ github.base_ref }}
+      test_groups: |
+        [
+          {
+            "npu_type":"cpu",
+            "runner_label":"linux-amd64-cpu-8-hk",
+            "test_paths":[
+              "tests/ut/_310p",
+              "tests/ut/_fake_weight",
+              "tests/ut/_tools",
+              "tests/ut/attention",
+              "tests/ut/compilation",
+              "tests/ut/core",
+              "tests/ut/device",
+              "tests/ut/device_allocator",
+              "tests/ut/distributed",
+              "tests/ut/eplb",
+              "tests/ut/helpers",
+              "tests/ut/kv_offload",
+              "tests/ut/lora",
+              "tests/ut/model_executor",
+              "tests/ut/model_loader",
+              "tests/ut/models",
+              "tests/ut/observability",
+              "tests/ut/ops",
+              "tests/ut/patch",
+              "tests/ut/profiler",
+              "tests/ut/quantization",
+              "tests/ut/sample",
+              "tests/ut/simple_kv_offload",
+              "tests/ut/spec_decode",
+              "tests/ut/tools",
+              "tests/ut/worker",
+              "tests/ut/xlite",
+              "tests/ut/test_ascend_config.py",
+              "tests/ut/test_ascend_forward_context.py",
+              "tests/ut/test_batch_invariant.py",
+              "tests/ut/test_compressed_prefix_cache.py",
+              "tests/ut/test_cpu_binding.py",
+              "tests/ut/test_dependency_documentation.py",
+              "tests/ut/test_envs.py",
+              "tests/ut/test_logger.py",
+              "tests/ut/test_meta_registration.py",
+              "tests/ut/test_platform.py",
+              "tests/ut/test_profiling_config.py",
+              "tests/ut/test_utils.py"
+            ]
+          }
+        ]
+
   recommend-tests-from-coverage:
     name: Recommend tests from coverage
     needs: pre-commit
@@ -494,13 +555,44 @@ jobs:
       base_ref: ${{ github.base_ref }}
       base_sha: ${{ needs.select-tests.outputs.base_sha }}
 
-  run-selected-tests:
+  get_latest_real_base_sha:
+    runs-on: ubuntu-latest
+    outputs:
+      real_base_sha: ${{ steps.fetch_main_sha.outputs.sha }}
+    steps:
+      - name: Fetch main branch latest commit hash remotely
+        id: fetch_main_sha
+        run: |
+          MAIN_SHA=$(git ls-remote --heads https://github.com/vllm-project/vllm-ascend.git main | awk '{print $1}')
+          echo "sha=${MAIN_SHA}" >> "$GITHUB_OUTPUT"
+
+  filter-test-groups:
+    needs: [select-tests]
+    if: ${{ !cancelled() && needs.select-tests.result == 'success' }}
+    runs-on: linux-amd64-cpu-8-hk
+    container:
+      image: quay.io/ascend/lint-base:latest
+    outputs:
+      npu_test_groups: ${{ steps.filter.outputs.npu_test_groups }}
+      cpu_test_groups: ${{ steps.filter.outputs.cpu_test_groups }}
+    steps:
+      - name: Filter test groups split cpu / npu
+        id: filter
+        run: |
+          set -euo pipefail
+          RAW=${{ needs.select-tests.outputs.test_groups }}
+          NPU_GROUPS=$(echo "${RAW}" | jq -c '[ .[] | select(.npu_type != "cpu") ]')
+          CPU_GROUPS=$(echo "${RAW}" | jq -c '[ .[] | select(.npu_type == "cpu") ]')
+          echo "npu_test_groups=${NPU_GROUPS}" >> "${GITHUB_OUTPUT}"
+          echo "cpu_test_groups=${CPU_GROUPS}" >> "${GITHUB_OUTPUT}"
+
+  run-selected-npu-tests:
     needs:
+      - get_latest_real_base_sha
       - pre-commit
       - select-tests
       - ensure-csrc-cache
-    # !cancelled() overrides transitive skip-propagation from recommend-tests-from-coverage
-    # (skipped under ready-all) through select-tests, without running after workflow cancel.
+      - filter-test-groups
     if: >-
       ${{
         !cancelled() &&
@@ -522,8 +614,40 @@ jobs:
       vllm: ${{ matrix.vllm_version }}
       ref: ${{ github.event.pull_request.head.sha }}
       base_ref: ${{ github.base_ref }}
-      base_sha: ${{ needs.select-tests.outputs.base_sha }}
-      test_groups: ${{ needs.select-tests.outputs.test_groups }}
+      base_sha: ${{ needs.get_latest_real_base_sha.outputs.real_base_sha }}
+      test_groups: ${{ needs.filter-test-groups.outputs.npu_test_groups }}
+    secrets: inherit
+
+  run-selected-dynamic-cpu-tests:
+    needs:
+      - get_latest_real_base_sha
+      - pre-commit
+      - select-tests
+      - ensure-csrc-cache
+      - filter-test-groups
+    if: >-
+      ${{
+        !cancelled() &&
+        needs.pre-commit.result == 'success' &&
+        needs.ensure-csrc-cache.result == 'success' &&
+        (
+          contains(github.event.pull_request.labels.*.name, 'ready-precise') ||
+          contains(github.event.pull_request.labels.*.name, 'ready-all')
+        ) &&
+        needs.select-tests.outputs.has_tests == 'true'
+      }}
+    strategy:
+      fail-fast: false
+      matrix:
+        vllm_version:
+          - ${{ needs.pre-commit.outputs.main_commit }}
+    uses: ./.github/workflows/_selected_tests_cpu.yaml
+    with:
+      vllm: ${{ matrix.vllm_version }}
+      ref: ${{ github.event.pull_request.head.sha }}
+      base_ref: ${{ github.base_ref }}
+      test_groups: ${{ needs.filter-test-groups.outputs.cpu_test_groups }}
+    secrets: inherit
 
   run-selected-tests-upstream:
     needs:
@@ -551,11 +675,10 @@ jobs:
       base_ref: ${{ github.base_ref }}
       base_sha: ${{ needs.select-tests.outputs.base_sha }}
       test_groups: ${{ needs.select-tests.outputs.test_groups }}
+    secrets: inherit
 
   run-selected-tests-a5:
     needs: [pre-commit, select-tests]
-    # !cancelled() overrides transitive skip-propagation from recommend-tests-from-coverage
-    # (skipped under ready-all) through select-tests, without running after workflow cancel.
     if: >-
       ${{
         !cancelled() &&
@@ -573,12 +696,14 @@ jobs:
       vllm: ${{ matrix.vllm_version }}
       ref: ${{ github.event.pull_request.head.sha }}
       base_ref: ${{ github.base_ref }}
+      base_sha: ${{ needs.select-tests.outputs.base_sha }}
       test_groups: ${{ needs.select-tests.outputs.test_groups_a5 }}
+    secrets: inherit
 
   # Summary gate job — use this single, stable name as the required status check
   # in GitHub branch protection rules instead of tracking individual matrix job names.
   ci-gate:
-    needs: [pre-commit, select-tests, run-selected-tests, run-selected-tests-a5]
+    needs: [pre-commit, select-tests, filter-test-groups, run-selected-npu-tests, run-selected-dynamic-cpu-tests, run-selected-tests-a5]
     if: ${{ !cancelled() }}
     runs-on: linux-amd64-cpu-2-hk
     steps:
@@ -624,7 +749,7 @@ jobs:
           fi
 
           if [ "${HAS_TESTS}" = "true" ]; then
-            TESTS_RESULT="${{ needs.run-selected-tests.result }}"
+            TESTS_RESULT="${{ needs.run-selected-npu-tests.result }}"
             echo "run-selected-tests: ${TESTS_RESULT}"
             if [ "${HAS_READY_LABEL}" != "true" ] && [ "${HAS_READY_ALL_LABEL}" != "true" ]; then
               echo "::error::Selected tests are required; add the ready-precise or ready-all label"