[v1] model loader (#9613)

tests/conftest.py

@@ -18,6 +18,7 @@ Contains shared fixtures, pytest configuration, and custom markers.
 """
 
 import pytest
+from pytest import Config, Item
 
 from llamafactory.extras.misc import get_current_device, is_env_enabled
 from llamafactory.train.test_utils import patch_valuehead_model
@@ -29,21 +30,15 @@ except Exception:
     CURRENT_DEVICE = "cpu"
 
 
-def pytest_configure(config):
+def pytest_configure(config: Config):
     """Register custom pytest markers."""
     config.addinivalue_line(
         "markers", "slow: marks tests as slow (deselect with '-m \"not slow\"' or set RUN_SLOW=1 to run)"
     )
-    config.addinivalue_line(
-        "markers", "skip_on_devices: skip test on specified devices, e.g., @pytest.mark.skip_on_devices('npu', 'xpu')"
-    )
-    config.addinivalue_line(
-        "markers", "require_device: test requires specific device, e.g., @pytest.mark.require_device('cuda')"
-    )
     config.addinivalue_line("markers", "runs_on: test requires specific device, e.g., @pytest.mark.runs_on(['cpu'])")
 
 
-def _handle_runs_on(items):
+def _handle_runs_on(items: list[Item]):
     """Skip tests on specified devices based on runs_on marker.
 
     Usage:
@@ -68,7 +63,7 @@ def _handle_runs_on(items):
                 )
 
 
-def _handle_slow_tests(items):
+def _handle_slow_tests(items: list[Item]):
     """Skip slow tests unless RUN_SLOW environment variable is set.
 
     Usage:
@@ -85,51 +80,9 @@ def _handle_slow_tests(items):
                 item.add_marker(skip_slow)
 
 
-def _handle_device_skips(items):
-    """Skip tests on specified devices based on skip_on_devices marker.
-
-    Usage:
-        @pytest.mark.skip_on_devices("npu", "xpu")
-        def test_something():
-            pass
-    """
-    for item in items:
-        skip_marker = item.get_closest_marker("skip_on_devices")
-        if skip_marker:
-            skip_devices = skip_marker.args
-            if CURRENT_DEVICE in skip_devices:
-                item.add_marker(
-                    pytest.mark.skip(
-                        reason=f"test skipped on {CURRENT_DEVICE.upper()} (skip list: {', '.join(skip_devices)})"
-                    )
-                )
-
-
-def _handle_device_requirements(items):
-    """Skip tests that require a specific device when running on other devices.
-
-    Usage:
-        @pytest.mark.require_device("cuda")
-        def test_gpu_only():
-            pass
-    """
-    for item in items:
-        require_marker = item.get_closest_marker("require_device")
-        if require_marker:
-            required_device = require_marker.args[0] if require_marker.args else None
-            if required_device and CURRENT_DEVICE != required_device:
-                item.add_marker(
-                    pytest.mark.skip(
-                        reason=f"test requires {required_device.upper()} (current: {CURRENT_DEVICE.upper()})"
-                    )
-                )
-
-
-def pytest_collection_modifyitems(config, items):
+def pytest_collection_modifyitems(config: Config, items: list[Item]):
     """Modify test collection based on markers and environment."""
     _handle_slow_tests(items)
-    _handle_device_skips(items)
-    _handle_device_requirements(items)
     _handle_runs_on(items)
 
 

tests/model/test_base.py

@@ -31,15 +31,13 @@ INFER_ARGS = {
 
 
 @pytest.mark.runs_on(["cpu", "npu"])
-@pytest.mark.skip_on_devices("npu")
 def test_base():
     model = load_infer_model(**INFER_ARGS)
     ref_model = load_reference_model(TINY_LLAMA3)
     compare_model(model, ref_model)
 
 
-@pytest.mark.runs_on(["cpu", "npu"])
-@pytest.mark.skip_on_devices("npu")
+@pytest.mark.runs_on(["cpu"])
 @pytest.mark.usefixtures("fix_valuehead_cpu_loading")
 def test_valuehead():
     model = load_infer_model(add_valuehead=True, **INFER_ARGS)

tests/model/test_lora.py

@@ -104,7 +104,6 @@ def test_lora_train_valuehead():
 
 
 @pytest.mark.runs_on(["cpu", "npu"])
-@pytest.mark.skip_on_devices("npu")
 def test_lora_inference():
     model = load_infer_model(**INFER_ARGS)
     ref_model = load_reference_model(TINY_LLAMA3, TINY_LLAMA_ADAPTER, use_lora=True).merge_and_unload()

tests/version.txt

@@ -1,2 +1,2 @@
 # change if test fails or cache is outdated
-0.9.4.103
+0.9.4.104