[Refactor] Reject unknown diffusion config fields (#5172)

Signed-off-by: TaffyOfficial <2324465096@qq.com>
Signed-off-by: zuiho <2324465096@qq.com>
Signed-off-by: Gao Han <hgaoaf@connect.ust.hk>
Co-authored-by: TaffyOfficial <2324465096@qq.com>
Co-authored-by: Gao Han <hgaoaf@connect.ust.hk>

tests/diffusion/quantization/test_quantization_fp8.py

@@ -414,18 +414,16 @@ def test_quantization_key_maps_to_quantization_config():
 
 
 @hardware_test(res={"cuda": ["L4", "B200"]})
-def test_quantization_config_key_takes_priority():
-    """When both 'quantization' and 'quantization_config' are set,
-    'quantization_config' takes priority."""
+def test_quantization_and_quantization_config_conflict():
+    """Two explicit quantization sources are rejected instead of silently merged."""
     from vllm_omni.diffusion.data import OmniDiffusionConfig
 
-    config = OmniDiffusionConfig.from_kwargs(
-        model="test",
-        quantization="fp8",
-        quantization_config={"method": "fp8", "activation_scheme": "static"},
-    )
-    assert config.quantization_config is not None
-    assert config.quantization_config.activation_scheme == "static"
+    with pytest.raises(ValueError, match=r"quantization.*quantization_config"):
+        OmniDiffusionConfig.from_kwargs(
+            model="test",
+            quantization="fp8",
+            quantization_config={"method": "fp8", "activation_scheme": "static"},
+        )
 
 
 @hardware_test(res={"cuda": ["L4", "B200"]})