[Refactor][OutputProcessor 2/8]: OmniRequestOutput should inherit RequestOutput and no nested wrap-up (#5146)

Signed-off-by: Boao Shi <aoibosh@connect.hku.hk>
Signed-off-by: bowieshi <aoibosh@connect.hku.hk>
Signed-off-by: princepride <wangzhipeng628@gmail.com>
Signed-off-by: R2-Y <ruiruyang2@gmail.com>
Signed-off-by: amy-why-3459 <wuhaiyan17@huawei.com>
Signed-off-by: Shi Boao Bowie <aoibosh@connect.hku.hk>
Co-authored-by: 汪志鹏 <wangzhipeng628@gmail.com>
Co-authored-by: ruirui(rein) yang <73573651+R2-Y@users.noreply.github.com>
Co-authored-by: amy-why-3459 <wuhaiyan17@huawei.com>
Co-authored-by: Gao Han <hgaoaf@connect.ust.hk>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
This commit is contained in:
authored and GitHub committed 2026-08-12 22:58:23 +08:00
1 parent 38a2500bdf
commit fcf1085395
89 files changed
+944 -639

No files matched your search

@@ -2156,12 +2156,12 @@ def test_direct_listen_decision_survives_inner_completion_metadata():
)
assert decision is not None
output = attach_duplex_output_decision(
OmniRequestOutput(
OmniRequestOutput.from_stage_output(
inner_output,
request_id="duplex-direct-listen",
finished=True,
stage_id=0,
final_output_type=decision.final_output_type,
request_output=inner_output,
metrics={
"stage_metrics": {
"0": {
@@ -3049,12 +3049,12 @@ async def test_minicpmo_auto_response_listen_without_response_does_not_defer_com
)
assert decision is not None
listen_output = attach_duplex_output_decision(
OmniRequestOutput(
OmniRequestOutput.from_stage_output(
inner_output,
request_id=request_id,
finished=True,
stage_id=0,
final_output_type=decision.final_output_type,
request_output=inner_output,
),
decision,
)
@@ -119,7 +119,6 @@ class MockGenerationResult:
def __init__(self, images):
self.images = images
self.request_output = SimpleNamespace(images=images)
self.stage_durations = {}
self.peak_memory_mb = 0.0
@@ -137,10 +136,7 @@ class MockStageResult:
outputs = [SimpleNamespace(text=text, index=0)]
else:
outputs = []
self.request_output = SimpleNamespace(
outputs=outputs,
images=self.images,
)
self.outputs = outputs
self.stage_durations = {}
self.peak_memory_mb = 0.0
@@ -2080,22 +2076,20 @@ def test_extract_images_from_result():
assert all(isinstance(img, Image.Image) for img in images)
assert all(img.size == (64, 64) for img in images)
# Test dict path: result.request_output["images"]
# Test result with "images" attribute set in __init__
class DictRequestOutput:
def __init__(self):
self.request_output = {"images": [np.random.randint(0, 255, (64, 64, 3), dtype=np.uint8)]}
self.images = [np.random.randint(0, 255, (64, 64, 3), dtype=np.uint8)]
result = DictRequestOutput()
images = _extract_images_from_result(result)
assert len(images) == 1
assert isinstance(images[0], Image.Image)
# Test attribute path: result.request_output.images
# Test result with "images" attribute from an inner object
class AttrRequestOutput:
def __init__(self):
self.request_output = type(
"obj", (), {"images": [np.random.randint(0, 255, (32, 32, 3), dtype=np.uint8)]}
)()
self.images = [np.random.randint(0, 255, (32, 32, 3), dtype=np.uint8)]
result = AttrRequestOutput()
images = _extract_images_from_result(result)
@@ -180,10 +180,10 @@ def create_mock_audio_output_for_test(
audio_tensor = torch.sin(torch.linspace(0, 440 * 2 * torch.pi, num_samples))
mock_request_output = MockRequestOutput(request_id=request_id, audio_tensor=audio_tensor)
return OmniRequestOutput(
return OmniRequestOutput.from_stage_output(
mock_request_output,
stage_id=0,
final_output_type="audio",
request_output=mock_request_output,
)
@@ -2292,10 +2292,10 @@ class TestStreamingResponse:
self.prompt_logprobs = None
self.kv_transfer_params = None
return OmniRequestOutput(
return OmniRequestOutput.from_stage_output(
MockRequestOutput(audio_tensor=chunk),
stage_id=0,
final_output_type="audio",
request_output=MockRequestOutput(audio_tensor=chunk),
finished=finished,
)
@@ -3441,10 +3441,10 @@ class TestWAVStreaming:
self.prompt_logprobs = None
self.kv_transfer_params = None
return OmniRequestOutput(
return OmniRequestOutput.from_stage_output(
MockRequestOutput(audio_tensor=chunk),
stage_id=0,
final_output_type="audio",
request_output=MockRequestOutput(audio_tensor=chunk),
finished=finished,
)
@@ -47,7 +47,7 @@ def _text_result(text: str) -> OmniRequestOutput:
output.text = text
request_output = RequestOutput()
request_output.outputs = [output]
return OmniRequestOutput(final_output_type="text", request_output=request_output)
return OmniRequestOutput.from_stage_output(request_output, final_output_type="text")
def _audio_result(audio_data: Any) -> OmniRequestOutput:
@@ -61,7 +61,7 @@ def _audio_result(audio_data: Any) -> OmniRequestOutput:
output.multimodal_output = {"audio": audio_data}
request_output = RequestOutput()
request_output.outputs = [output]
return OmniRequestOutput(final_output_type="audio", request_output=request_output)
return OmniRequestOutput.from_stage_output(request_output, final_output_type="audio")
class MockWebSocket:
+4 -6
View File
@@ -480,11 +480,9 @@ def test_async_omni_duplex_direct_output_prefers_outer_control_metadata():
OmniRequestOutput(
request_id="duplex-sid-e0-stage0",
stage_id=0,
request_output=SimpleNamespace(
outputs=[],
multimodal_output=SimpleNamespace(kind="processed-payload"),
_custom_output={"inner": "completion"},
),
outputs=[],
_multimodal_output={"kind": "processed-payload"},
_custom_output={"inner": "completion"},
),
decision,
)
@@ -557,7 +555,7 @@ async def test_async_omni_duplex_collect_wraps_raw_response_stage_output():
assert outputs[0].finished is False
assert outputs[0].stage_id == 1
assert outputs[0].final_output_type == "audio"
assert outputs[0].request_output is raw_stage1_output
assert outputs[0].peak_memory_mb == raw_stage1_output.peak_memory_mb
def test_async_omni_duplex_request_info_includes_response_stage():
+9 -10
View File
@@ -72,8 +72,8 @@ def make_output_msg(
final_output_type=final_output_type,
images=images or [],
stage_durations={},
outputs=[SimpleNamespace(text=payload, index=0)],
)
engine_output.payload = payload
return OutputMessage(
request_id=request_id,
stage_id=stage_id,
@@ -474,7 +474,7 @@ async def test_async_omni_yields_only_final_stage_outputs(monkeypatch: pytest.Mo
app.shutdown()
assert [output.stage_id for output in outputs] == [2]
assert [output.request_output.payload for output in outputs] == ["final"]
assert [output.outputs[0].text for output in outputs] == ["final"]
assert "req-1" not in app.request_states
@@ -492,7 +492,7 @@ async def test_async_omni_accepts_multiple_final_stage_streams(monkeypatch: pyte
app.shutdown()
assert [output.stage_id for output in outputs] == [0, 0, 0, 2, 2, 2]
assert [output.request_output.payload for output in outputs] == [
assert [output.outputs[0].text for output in outputs] == [
"req-1-stage0-0",
"req-1-stage0-1",
"req-1-stage0-2",
@@ -517,7 +517,7 @@ async def test_async_omni_stops_on_final_stage_finished(monkeypatch: pytest.Monk
finally:
app.shutdown()
assert [output.request_output.payload for output in outputs] == [
assert [output.outputs[0].text for output in outputs] == [
"req-1-stage0",
"req-1-stage2-final",
]
@@ -544,7 +544,7 @@ async def test_async_omni_diffusion_only_yields_single_image_output(monkeypatch:
assert outputs[0].stage_id == 0
assert outputs[0].final_output_type == "image"
assert outputs[0].images == ["req-1-image"]
assert outputs[0].request_output.payload == "req-1-diffusion-final"
assert outputs[0].outputs[0].text == "req-1-diffusion-final"
@pytest.mark.asyncio
@@ -565,7 +565,7 @@ async def test_async_omni_llm_diffusion_yields_text_stream_then_image(monkeypatc
assert [output.stage_id for output in outputs] == [0, 0, 0, 1]
assert [output.final_output_type for output in outputs] == ["text", "text", "text", "image"]
assert [output.request_output.payload for output in outputs] == [
assert [output.outputs[0].text for output in outputs] == [
"req-1-text-0",
"req-1-text-1",
"req-1-text-2",
@@ -695,7 +695,7 @@ def test_omni_generate_py_generator_yields_final_outputs_for_each_request(monkey
assert len(outputs) == 4
assert [output.stage_id for output in outputs] == [0, 2, 0, 2]
assert [output.request_output.payload for output in outputs] == [
assert [output.outputs[0].text for output in outputs] == [
f"{engine.submitted[0]['request_id']}-stage0-0",
f"{engine.submitted[0]['request_id']}-stage2-final",
f"{engine.submitted[1]['request_id']}-stage0-0",
@@ -739,7 +739,7 @@ def test_omni_generate_diffusion_only_yields_single_image_per_request(monkeypatc
assert len(outputs) == 2
assert [output.stage_id for output in outputs] == [0, 0]
assert [output.final_output_type for output in outputs] == ["image", "image"]
assert [output.request_output.payload for output in outputs] == [
assert [output.outputs[0].text for output in outputs] == [
f"{engine.submitted[0]['request_id']}-diffusion-final",
f"{engine.submitted[1]['request_id']}-diffusion-final",
]
@@ -767,7 +767,7 @@ def test_omni_generate_llm_diffusion_yields_final_text_then_image_per_request(
assert len(outputs) == 4
assert [output.stage_id for output in outputs] == [0, 1, 0, 1]
assert [output.final_output_type for output in outputs] == ["text", "image", "text", "image"]
assert [output.request_output.payload for output in outputs] == [
assert [output.outputs[0].text for output in outputs] == [
f"{engine.submitted[0]['request_id']}-text-0",
f"{engine.submitted[0]['request_id']}-image-final",
f"{engine.submitted[1]['request_id']}-text-0",
@@ -1046,7 +1046,6 @@ def _enqueue_stage_error(
if kill_engine:
engine._alive = False
engine_output = OmniRequestOutput.from_error(msg["request_id"], error_text)
engine_output.payload = ""
engine.output_q.put_nowait(
OutputMessage(
request_id=msg["request_id"],
@@ -69,10 +69,10 @@ def _make_text_omni_output(
],
finished=finish_reason is not None,
)
return OmniRequestOutput(
return OmniRequestOutput.from_stage_output(
res,
request_id=request_id,
final_output_type="text",
request_output=res,
finished=finish_reason is not None,
)
@@ -108,12 +108,12 @@ def _make_audio_omni_output(
outputs=[completion],
finished=True,
)
return OmniRequestOutput(
return OmniRequestOutput.from_stage_output(
res,
request_id=request_id,
stage_id=stage_id,
replica_id=replica_id,
final_output_type="audio",
request_output=res,
finished=True,
)
@@ -151,7 +151,7 @@ def _build_serving_chat():
)
instance._create_audio_choice = MagicMock(
side_effect=lambda omni_res, role, request, stream=False: _mock_audio_choices(
index=omni_res.request_output.outputs[0].index,
index=omni_res.outputs[0].index,
role=role,
)
)
@@ -489,7 +489,7 @@ async def test_audio_chunk_without_waveform_keeps_stream_alive():
request = _make_request(modalities=["text", "audio"])
empty_audio = _make_audio_omni_output()
empty_audio.request_output.outputs[0].multimodal_output = {"audio": []}
empty_audio.outputs[0].multimodal_output = {"audio": []}
def create_audio_choice(omni_res, role, request, stream=False):
return OmniOpenAIServingChat._create_audio_choice(serving_chat, omni_res, role, request, stream=stream)