[CI] run tests/compile/test_config.py

.buildkite/test-pipeline.yaml

@@ -375,6 +375,7 @@ steps:
     - tests/compile
   commands:
     - pytest -v -s compile/test_pass_manager.py
+    - pytest -v -s compile/test_config.py
     - pytest -v -s compile/test_fusion.py
     - pytest -v -s compile/test_fusion_attn.py
     - pytest -v -s compile/test_silu_mul_quant_fusion.py

tests/compile/test_config.py

@@ -4,7 +4,8 @@
 
 import vllm
 from vllm.compilation.counter import compilation_counter
-from vllm.config import CompilationConfig, VllmConfig
+from vllm.config import (CompilationConfig, CUDAGraphMode, ModelConfig,
+                         VllmConfig)
 from vllm.utils import _is_torch_equal_or_newer
 
 
@@ -18,12 +19,35 @@ def test_version():
 
 def test_use_cudagraphs_dynamic(monkeypatch):
     assert vllm.envs.VLLM_USE_V1
-    vllm_config = VllmConfig()
-    assert vllm_config.compilation_config.use_cudagraph
+    # cudagraph_mode=PIECEWISE by default
+    config = VllmConfig()
+    assert config.compilation_config.cudagraph_mode == CUDAGraphMode.PIECEWISE
+
+    config2 = VllmConfig(compilation_config=CompilationConfig(
+        cudagraph_mode=CUDAGraphMode.NONE))
+    config3 = VllmConfig(compilation_config=CompilationConfig(
+        cudagraph_mode=CUDAGraphMode.FULL))
+    config4 = VllmConfig(model_config=ModelConfig(enforce_eager=True))
+    assert config2.compilation_config.cudagraph_mode == CUDAGraphMode.NONE
+    assert config3.compilation_config.cudagraph_mode == CUDAGraphMode.FULL
+    assert config4.compilation_config.cudagraph_mode == CUDAGraphMode.NONE
 
     monkeypatch.setenv('VLLM_USE_V1', '0')
-    vllm_config = VllmConfig()
-    assert not vllm_config.compilation_config.use_cudagraph
+    # TODO remove when V0 fully removed:
+    # https://github.com/vllm-project/vllm/issues/18571
+
+    config = VllmConfig()  # cudagraph_mode=NONE by default
+    assert config.compilation_config.cudagraph_mode == CUDAGraphMode.NONE
+
+    config2 = VllmConfig(compilation_config=CompilationConfig(
+        cudagraph_mode=CUDAGraphMode.PIECEWISE))
+    config3 = VllmConfig(compilation_config=CompilationConfig(
+        cudagraph_mode=CUDAGraphMode.FULL))
+    config4 = VllmConfig(model_config=ModelConfig(enforce_eager=True))
+
+    assert config2.compilation_config.cudagraph_mode == CUDAGraphMode.PIECEWISE
+    assert config3.compilation_config.cudagraph_mode == CUDAGraphMode.FULL
+    assert config4.compilation_config.cudagraph_mode == CUDAGraphMode.NONE
 
 
 def test_custom_op():

vllm/config/__init__.py

@@ -483,8 +483,8 @@ def __post_init__(self):
         # settings (see the below code).
         if self.compilation_config.level is None:
             if envs.VLLM_USE_V1:
-                if (self.model_config is not None
-                        and not self.model_config.enforce_eager):
+                if self.model_config is None or \
+                        not self.model_config.enforce_eager:
                     self.compilation_config.level = CompilationLevel.PIECEWISE
                 else:
                     self.compilation_config.level = \