Remove deprecated fields from CompilationConfig (#27593)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

tests/compile/piecewise/test_multiple_graphs.py

@@ -203,7 +203,7 @@ def test_multi_graph_piecewise_compile(use_inductor_graph_partition: bool):
     vllm_config = VllmConfig(
         compilation_config=CompilationConfig(
             mode=CompilationMode.VLLM_COMPILE,
-            use_cudagraph=True,
+            cudagraph_mode=CUDAGraphMode.PIECEWISE,
             splitting_ops=["silly::attention"],
             cudagraph_capture_sizes=[1, 2],
             use_inductor_graph_partition=use_inductor_graph_partition,
@@ -281,7 +281,7 @@ def test_multi_graph_piecewise_compile(use_inductor_graph_partition: bool):
     vllm_config = VllmConfig(
         compilation_config=CompilationConfig(
             mode=CompilationMode.VLLM_COMPILE,
-            use_cudagraph=False,
+            cudagraph_mode=CUDAGraphMode.NONE,
             splitting_ops=["silly::attention"],
             use_inductor_graph_partition=use_inductor_graph_partition,
         )

tests/compile/piecewise/test_simple.py

@@ -62,7 +62,6 @@ def _run_simple_model(
     vllm_config = VllmConfig(
         compilation_config=CompilationConfig(
             mode=CompilationMode.VLLM_COMPILE,
-            use_cudagraph=True,
             use_inductor=use_inductor,
             splitting_ops=splitting_ops,
             use_inductor_graph_partition=use_inductor_graph_partition,

tests/compile/piecewise/test_toy_llama.py

@@ -449,7 +449,6 @@ def benchmark():
         if piecewise:
             compilation_config = CompilationConfig(
                 mode=CompilationMode.VLLM_COMPILE,
-                use_cudagraph=True,
                 splitting_ops=["silly::attention"],
                 cudagraph_capture_sizes=cudagraph_sizes,
             )