[CI] allow CI skip test & clean ci logic (#2771)

* [CI] add skip to allow CI skip it

* [CI] merge xpu check into no gpu check step

* [CI] fix KeyError: 'group not found: tests/models/awq'

Author: CSY-ModelCloud

.github/scripts/ci_workflow.py

@@ -89,7 +89,10 @@ def parse_test_config_for_path(
         scoped_test_name = None
         if index == len(candidate_groups) - 1:
             scoped_test_name = leaf_name
-        scoped = parse_test_config(yaml_file, candidate_group, scoped_test_name)
+        try:
+            scoped = parse_test_config(yaml_file, candidate_group, scoped_test_name)
+        except KeyError:
+            continue
         result.update(scoped)
     return result
 
@@ -125,17 +128,33 @@ def is_model_compat_test(rel_path: str, file_path: Path) -> bool:
     return any(marker in contents for marker in markers)
 
 
+def matches_test_regex(test_regex: str, rel_path: str) -> bool:
+    if re.match(test_regex, rel_path):
+        return True
+    return re.match(test_regex, PurePosixPath(rel_path).name) is not None
+
+
+def should_skip_test(yaml_file: str | Path, rel_path: str) -> bool:
+    try:
+        config = parse_test_config_for_path(yaml_file, "tests", rel_path)
+    except KeyError:
+        return False
+    return bool(config.get("skip", False))
+
+
 def list_tests(ignored_test_files: str | list[str], test_names: str, test_regex: str, tests_root: str | Path) -> tuple[list[str], list[str], list[str]]:
     tests_root = Path(tests_root)
     input_tests = [strip_py_suffix(name) for name in split_csv(test_names)]
     ignored_raw = ignored_test_files if isinstance(ignored_test_files, list) else split_csv(ignored_test_files)
     ignored_set = {strip_py_suffix(name) for name in ignored_raw}
+    yaml_file = Path(__file__).with_name("test.yaml")
 
     all_tests = {
         rel: path
         for path in tests_root.rglob("test_*.py")
         for rel in [str(path.relative_to(tests_root).with_suffix(""))]
         if rel not in ignored_set and path.stem not in ignored_set
+        if not should_skip_test(yaml_file, rel)
     }
 
     model_tests = {
@@ -145,7 +164,7 @@ def list_tests(ignored_test_files: str | list[str], test_names: str, test_regex:
            and "mlx" not in rel
            and "ipex" not in rel
            and "xpu" not in rel
-           and re.match(test_regex, rel)
+           and matches_test_regex(test_regex, rel)
            and is_model_compat_test(rel, path)
     }
 
@@ -157,15 +176,15 @@ def list_tests(ignored_test_files: str | list[str], test_names: str, test_regex:
            and "mlx" not in rel
            and "ipex" not in rel
            and "xpu" not in rel
-           and re.match(test_regex, rel)
+           and matches_test_regex(test_regex, rel)
     }
 
     mlx_tests = {
         rel
         for rel in all_tests
         if ("mlx" in rel or "apple" in rel)
            and ((rel in input_tests) if input_tests else True)
-           and re.match(test_regex, rel)
+           and matches_test_regex(test_regex, rel)
     }
 
     return (

.github/scripts/test.yaml

@@ -2,6 +2,7 @@ common:
   py: 3.14t # common config
   gpu: 1
   sm: 0
+  skip: false
 
 tests:
   test_multi_gpu_inference: # per-test config
@@ -18,4 +19,5 @@ tests/models:
 
 tests/kernels:
   test_asymmetric_real_models:
-    sm: '9.0'
+    sm: '9.0'
+    skip: true # no SM90 card

.github/workflows/unit_tests.yml

@@ -301,37 +301,6 @@ jobs:
           echo "--- uninstalling required deps..."
           python .github/scripts/ci_deps.py uninstall ${{ matrix.test_script }}
 
-      # - name: Install Evalution
-      #   run: |
-      #     uv pip install git+https://x-access-token:${{ secrets.REPO_TOKEN }}@github.com/ModelCloud/Evalution.git
-      #      - name: Install requirements
-      #        run: |
-      #          bash -c "$(curl -L http://${RUNNER}/scripts/env/init_compiler_no_env.sh)" @ ${{ needs.check-vm.outputs.cuda_version }} ${{ env.TORCH_VERSION }} $python_version${{ env.PYTHON_VERSION }}
-
-      #      - name: Download source from local
-      #        continue-on-error: true
-      #        run: |
-      #          curl -s -O  http://$RUNNER/whl/${{ env.repo }}/${{ github.run_id }}/gptqmodel_source.tar.gz
-      #          ls -ahl .
-      #          sha256=$(sha256sum $file_name)
-      #          echo "sha256=$sha256"
-      #          echo "SOURCE_DOWNLOADED=1" >> $GITHUB_ENV
-
-      # - name: Download source from github
-      #   if: env.SOURCE_DOWNLOADED == '' && !cancelled()
-      #   uses: actions/download-artifact@v8
-      #   with:
-      #     name: source
-      #     path: dist
-      #     run-id: ${{ github.run_id }}
-
-      #     - name: Uncompress source
-      #      continue-on-error: true
-      #     run: |
-      #      find . -mindepth 1 ! -name "gptqmodel_source.tar.gz" -exec rm -rf {} +
-      #     ls -ahl .
-      #    tar -zxf gptqmodel_source.tar.gz
-
       - name: Install package from source
         run: |
           uv pip uninstall gptqmodel || true
@@ -394,12 +363,14 @@ jobs:
         run: |
           if grep -q '^# GPU=-1$' "${{ matrix.test_script }}"; then
             echo "SKIP_GPU_ALLOCATION=true" >> "$GITHUB_ENV"
+          elif [[ "${{ matrix.test_script }}" == *xpu* ]]; then
+            echo "SKIP_GPU_ALLOCATION=true" >> "$GITHUB_ENV"
           else
             echo "SKIP_GPU_ALLOCATION=false" >> "$GITHUB_ENV"
           fi
 
       - name: Find suitable GPU
-        if: ${{ !contains(matrix.test_script, 'ipex') && !contains(matrix.test_script, 'xpu') && !cancelled() && env.SKIP_GPU_ALLOCATION != 'true' }}
+        if: ${{ !cancelled() && env.SKIP_GPU_ALLOCATION != 'true' }}
         run: |
           python .github/scripts/ci_gpu.py allocate \
             --base-url "${GPU_ALLOCATOR_URL}" \
@@ -413,9 +384,6 @@ jobs:
       - name: Run tests
         run: |
           extra_args=()
-          if [[ "${{ matrix.test_script }}" == *ipex* ]]; then
-            extra_args+=(--clear-cuda)
-          fi
           if [[ "${{ matrix.test_script }}" == *xpu* ]]; then
             extra_args+=(--clear-cuda --xpu-mode)
           fi
@@ -436,7 +404,7 @@ jobs:
             --test-script "${{ matrix.test_script }}"
 
       - name: Release GPU
-        if: always() && !contains(matrix.test_script, 'ipex') && !contains(matrix.test_script, 'xpu') && env.SKIP_GPU_ALLOCATION != 'true'
+        if: always() && env.SKIP_GPU_ALLOCATION != 'true'
         run: |
           python .github/scripts/ci_gpu.py release \
             --base-url "${GPU_ALLOCATOR_URL}" \
@@ -628,6 +596,8 @@ jobs:
         run: |
           if grep -q '^# GPU=-1$' "${{ matrix.test_script }}"; then
             echo "SKIP_GPU_ALLOCATION=true" >> "$GITHUB_ENV"
+          elif [[ "${{ matrix.test_script }}" == *xpu* ]]; then
+            echo "SKIP_GPU_ALLOCATION=true" >> "$GITHUB_ENV"
           else
             echo "SKIP_GPU_ALLOCATION=false" >> "$GITHUB_ENV"
           fi