[compile] raise on compile_size implicit padding (#32343)

Signed-off-by: dolpm <34420038+dolpm@users.noreply.github.com>

tests/compile/test_config.py

@@ -470,3 +470,68 @@ def test_cached_compilation_config(default_vllm_config):
 
     code = " ".join(code)
     assert "torch.ops._C.static_scaled_fp8_quant.default(" in code
+
+
+def test_compile_sizes_padding_validation():
+    """Test that compile_sizes with values that would be padded raises an error."""
+    # cudagraph_capture_sizes=[1, 2, 4, 8] means:
+    # - size 1 -> padded to 1
+    # - size 2 -> padded to 2
+    # - size 3 -> padded to 4
+    # - size 4 -> padded to 4
+    # - size 5 -> padded to 8
+    # etc.
+    # So compile_sizes=[3] should fail because 3 would be padded to 4
+
+    with pytest.raises(ValueError, match="would be padded to"):
+        config = CompilationConfig(
+            cudagraph_capture_sizes=[1, 2, 4, 8],
+            max_cudagraph_capture_size=8,
+            compile_sizes=[3],
+        )
+        config.post_init_cudagraph_sizes()
+
+    with pytest.raises(ValueError, match="would be padded to"):
+        config = CompilationConfig(
+            cudagraph_capture_sizes=[1, 2, 4, 8],
+            max_cudagraph_capture_size=8,
+            compile_sizes=[5],
+        )
+        config.post_init_cudagraph_sizes()
+
+    config = CompilationConfig(
+        cudagraph_capture_sizes=[1, 2, 4, 8],
+        max_cudagraph_capture_size=8,
+        compile_sizes=[1, 2, 4, 8],
+    )
+    config.post_init_cudagraph_sizes()
+    assert sorted(config.compile_sizes) == [1, 2, 4, 8]
+
+    config = CompilationConfig(
+        cudagraph_capture_sizes=[1, 2, 4, 8],
+        max_cudagraph_capture_size=8,
+        compile_sizes=["cudagraph_capture_sizes"],
+    )
+    config.post_init_cudagraph_sizes()
+    assert sorted(config.compile_sizes) == [1, 2, 4, 8]
+
+    # When cudagraphs are disabled (max_cudagraph_capture_size=0),
+    # padding validation should be skipped
+    config = CompilationConfig(
+        cudagraph_capture_sizes=[],
+        max_cudagraph_capture_size=0,
+        compile_sizes=[3, 5, 7],  # would be invalid with cudagraphs
+    )
+    config.post_init_cudagraph_sizes()
+    assert sorted(config.compile_sizes) == [3, 5, 7]
+
+    # When cudagraph_mode is NONE but capture_sizes is non-empty,
+    # padding validation should still be skipped
+    config = CompilationConfig(
+        cudagraph_capture_sizes=[1, 2, 4, 8],
+        max_cudagraph_capture_size=8,
+        cudagraph_mode=CUDAGraphMode.NONE,
+        compile_sizes=[3, 5, 7],  # would be invalid if cudagraphs were enabled
+    )
+    config.post_init_cudagraph_sizes()
+    assert sorted(config.compile_sizes) == [3, 5, 7]

vllm/config/compilation.py

@@ -909,6 +909,20 @@ class CompilationConfig:
         # May get recomputed in the model runner if adjustment is needed for spec-decode
         self.compute_bs_to_padded_graph_size()
 
+        # Validate that compile_sizes won't be changed by padding.
+        # Only validate when cudagraphs are actually being used.
+        if self.compile_sizes and self.cudagraph_mode != CUDAGraphMode.NONE:
+            for size in self.compile_sizes:
+                if size <= self.max_cudagraph_capture_size:
+                    padded = self.bs_to_padded_graph_size[size]
+                    if padded != size:
+                        raise ValueError(
+                            f"compile_sizes contains {size} which would be "
+                            f"padded to {padded}. All compile_sizes must be "
+                            "values that won't be changed by cudagraph padding. "
+                            "Use values from cudagraph_capture_sizes."
+                        )
+
     def set_splitting_ops_for_v1(
         self, all2all_backend: str, data_parallel_size: int = 1
     ):