[Feature] Support text encoder Disaggregation for H3 (#5885)

Signed-off-by: Canlin Guo <canlinguosdu@gmail.com>
Signed-off-by: lishunyang12 <lishunyang12@163.com>
Signed-off-by: yuanwu <yuan.wu@intel.com>
Co-authored-by: lishunyang12 <lishunyang12@163.com>
Co-authored-by: yuanwu <yuan.wu@intel.com>

tests/entrypoints/openai_api/test_serving_video_output_stream.py

@@ -47,7 +47,13 @@ def _build_test_app(
     engine_client.abort = mocker.AsyncMock()
     engine_client.submit_interaction_async = mocker.AsyncMock()
     engine_client.default_sampling_params_list = [OmniDiffusionSamplingParams()]
-    engine_client.stage_configs = [SimpleNamespace(stage_type="diffusion")]
+    engine_client.stage_configs = [
+        SimpleNamespace(
+            stage_type="diffusion",
+            final_output=True,
+            final_output_type="video",
+        )
+    ]
     engine_client.generate = mock_generate
 
     class FakeStreamingVideoEncoder:
@@ -72,10 +78,6 @@ def _build_test_app(
         "vllm_omni.entrypoints.openai.serving_video_output_stream.create_streaming_video_encoder",
         encoder_factory,
     )
-    mocker.patch(
-        "vllm_omni.entrypoints.openai.serving_video_output_stream.get_stage_type",
-        return_value="diffusion",
-    )
     mocker.patch(
         "vllm_omni.entrypoints.openai.serving_video_output_stream.build_stage_sampling_params_list",
         return_value=[OmniDiffusionSamplingParams()],
@@ -398,7 +400,13 @@ def _build_handler_for_async_tests(
     engine_client.abort = mocker.AsyncMock()
     engine_client.submit_interaction_async = mocker.AsyncMock()
     engine_client.default_sampling_params_list = [OmniDiffusionSamplingParams()]
-    engine_client.stage_configs = [SimpleNamespace(stage_type="diffusion")]
+    engine_client.stage_configs = [
+        SimpleNamespace(
+            stage_type="diffusion",
+            final_output=True,
+            final_output_type="video",
+        )
+    ]
     engine_client.generate = mock_generate
 
     chunks = streaming_chunks or [(b"mp4-chunk-0", True)]
@@ -421,10 +429,6 @@ def _build_handler_for_async_tests(
         "vllm_omni.entrypoints.openai.serving_video_output_stream.create_streaming_video_encoder",
         encoder_factory,
     )
-    mocker.patch(
-        "vllm_omni.entrypoints.openai.serving_video_output_stream.get_stage_type",
-        return_value="diffusion",
-    )
     mocker.patch(
         "vllm_omni.entrypoints.openai.serving_video_output_stream.build_stage_sampling_params_list",
         return_value=[OmniDiffusionSamplingParams()],

tests/entrypoints/openai_api/test_video_pipeline_capability.py

@@ -0,0 +1,66 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
+from types import SimpleNamespace
+
+import pytest
+
+from vllm_omni.diffusion.io_support import get_diffusion_output_type
+from vllm_omni.entrypoints.openai.utils import is_video_generation_pipeline
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
+
+
+def test_video_pipeline_requires_declared_final_video_stage():
+    assert is_video_generation_pipeline(
+        [
+            SimpleNamespace(
+                stage_type="llm",
+                final_output=False,
+                final_output_type=None,
+            ),
+            SimpleNamespace(
+                stage_type="diffusion",
+                final_output=True,
+                final_output_type="video",
+            ),
+        ]
+    )
+
+
+@pytest.mark.parametrize(
+    "stage_configs",
+    [
+        [SimpleNamespace(stage_type="diffusion")],
+        [
+            SimpleNamespace(
+                stage_type="diffusion",
+                final_output=True,
+                final_output_type="image",
+            )
+        ],
+        [
+            {
+                "stage_type": "diffusion",
+                "final_output": False,
+                "final_output_type": "video",
+            }
+        ],
+    ],
+)
+def test_video_pipeline_rejects_non_video_final_outputs(stage_configs):
+    assert not is_video_generation_pipeline(stage_configs)
+
+
+@pytest.mark.parametrize(
+    "model_class_name",
+    [
+        "LTX2TwoStagePipeline",
+        "LTX2DistilledOneStagePipeline",
+        "WanDMDPipeline",
+        "LingBotWorldCausalDMDPipeline",
+        "LongCatVideoAvatarPipeline",
+    ],
+)
+def test_registered_video_aliases_declare_video_output(model_class_name):
+    assert get_diffusion_output_type(model_class_name) == "video"

tests/entrypoints/openai_api/test_video_server.py

@@ -64,7 +64,13 @@ class MockVideoResult:
 
 class FakeAsyncOmni:
     def __init__(self):
-        self.stage_configs = [SimpleNamespace(stage_type="diffusion")]
+        self.stage_configs = [
+            SimpleNamespace(
+                stage_type="diffusion",
+                final_output=True,
+                final_output_type="video",
+            )
+        ]
         self.default_sampling_params_list = [OmniDiffusionSamplingParams()]
         self.model_class_name = "WanPipeline"
         self.captured_prompt = None
@@ -828,6 +834,18 @@ def test_multi_video_generation_preserves_uploaded_files_until_generation(
     assert engine.captured_sampling_params_list[0].extra_args["duration"] == 15.0
 
 
+def test_mixed_reference_capability_uses_model_metadata_when_config_defaults_false(test_client):
+    handler = test_client.app.state.openai_serving_video
+    handler._engine_client.model_class_name = None
+    handler._engine_client.stage_configs = [
+        SimpleNamespace(engine_args={"model_class_name": "MiniMaxH3TextEncoder"}),
+        SimpleNamespace(engine_args={"model_class_name": "MiniMaxH3Pipeline"}),
+    ]
+    handler._stage_configs = handler._engine_client.stage_configs
+
+    assert handler.supports_mixed_reference_inputs
+
+
 def test_decode_video_bytes_can_keep_first_frames():
     from vllm_omni.entrypoints.openai.video_api_utils import _decode_video_bytes