[Model] Reduce MOSS-TTS Local intermediate output overhead

Keep internal talker control outputs FINAL_ONLY while codec audio remains incremental. Reapply pipeline sampling constraints after default streaming coercion, log routine coercions at debug level, and cover explicit/default parameters with regression tests. Document matched high-concurrency statistics settings.

Signed-off-by: Sy03 <1370724210@qq.com>

examples/online_serving/text_to_speech/README.md

@@ -432,13 +432,19 @@ For a single H200, the optional
 deployment places both the talker and codec on logical GPU 0:
 
 ```bash
-CUDA_VISIBLE_DEVICES=0 vllm serve OpenMOSS-Team/MOSS-TTS-Local-Transformer-v1.5 \
+CUDA_VISIBLE_DEVICES=0 VLLM_OMNI_EVENT_DRIVEN_ORCH=1 \
+    vllm serve OpenMOSS-Team/MOSS-TTS-Local-Transformer-v1.5 \
     --omni --trust-remote-code \
-    --deploy-config vllm_omni/deploy/moss_tts_local_h200.yaml
+    --deploy-config vllm_omni/deploy/moss_tts_local_h200.yaml \
+    --disable-log-stats
 ```
 
 Run from the repository root and select an available physical GPU with
-`CUDA_VISIBLE_DEVICES`. This preset configures 256 request slots per stage,
+`CUDA_VISIBLE_DEVICES`. The command enables event-driven orchestration and
+disables detailed per-request statistics logging to reduce CPU overhead at
+high concurrency. Remove `--disable-log-stats` when those statistics are needed;
+keep these settings identical when comparing performance.
+This preset configures 256 request slots per stage,
 a 32 GiB talker KV cache, talker CUDA Graph buckets through 512 scheduled tokens,
 and codec batch buckets through 256. It requires more than 80 GiB of GPU memory;
 request capacity also depends on input and generated lengths. The default

tests/entrypoints/test_omni_entrypoints.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 from __future__ import annotations
 
 import queue
@@ -97,9 +100,11 @@ class FakeAsyncOmniEngine:
     ) -> None:
         self.model = model
         self.config_path = None
-        self.stage_configs: list[Any] = []
         self.stage_metadata = stage_metadata or [THREE_STAGE_META[-1]]
         self.num_stages = len(self.stage_metadata)
+        self.stage_configs = [
+            StageConfig(stage_id=i, model_stage="dummy-model").to_omegaconf() for i in range(self.num_stages)
+        ]
         self.default_sampling_params_list = default_sampling_params_list or [
             SamplingParams(max_tokens=8) for _ in range(self.num_stages)
         ]

tests/model_executor/models/moss_tts/test_local_output_policy.py

@@ -0,0 +1,37 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
+import pytest
+from vllm import SamplingParams
+from vllm.sampling_params import RequestOutputKind
+
+from vllm_omni.entrypoints.omni_base import OmniBase
+from vllm_omni.entrypoints.utils import coerce_param_message_types
+from vllm_omni.model_executor.models.moss_tts.pipeline import MOSS_TTS_LOCAL_PIPELINE
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
+
+
+@pytest.mark.parametrize("use_defaults", [False, True])
+def test_local_talker_control_outputs_preserve_codec_streaming(mocker, use_defaults):
+    base = object.__new__(OmniBase)
+    base.engine = mocker.Mock(num_stages=2)
+    base.sampling_constraints_list = [dict(stage.sampling_constraints) for stage in MOSS_TTS_LOCAL_PIPELINE.stages]
+    base.default_sampling_params_list = [
+        base._apply_sampling_constraints(SamplingParams(max_tokens=8, seed=11), constraints)
+        for constraints in base.sampling_constraints_list
+    ]
+    caller = coerce_param_message_types(
+        [SamplingParams(max_tokens=8, seed=11), SamplingParams(max_tokens=8, seed=11)], True
+    )
+    result = base.resolve_sampling_params_list(None if use_defaults else caller, allow_delta_coercion=True)
+
+    assert result[0].output_kind == RequestOutputKind.FINAL_ONLY
+    assert result[0].detokenize is False
+    assert result[0].max_tokens == 8
+    assert result[0].seed == 11
+    assert result[1].output_kind == RequestOutputKind.DELTA
+    assert caller[0].output_kind == RequestOutputKind.DELTA
+    assert base.default_sampling_params_list[0].output_kind == RequestOutputKind.FINAL_ONLY
+    assert MOSS_TTS_LOCAL_PIPELINE.stages[0].final_output is False
+    assert MOSS_TTS_LOCAL_PIPELINE.stages[1].final_output_type == "audio"

vllm_omni/entrypoints/omni_base.py

@@ -354,7 +354,9 @@ class OmniBase(PDDisaggregationMixin):
         if len(normalized) != self.num_stages:
             raise ValueError(f"Expected {self.num_stages} sampling params, got {len(normalized)}")
 
-        if sampling_params_list is not None:
+        # Streaming coercion may also change a constrained output kind in
+        # the defaults (for example, an internal TTS stage's FINAL_ONLY).
+        if sampling_params_list is not None or allow_delta_coercion:
             normalized = [
                 self._apply_sampling_constraints(params, constraints)
                 for params, constraints in zip(normalized, self.sampling_constraints_list, strict=True)

vllm_omni/entrypoints/utils.py

@@ -868,9 +868,9 @@ def maybe_coerce_to_message_type(params: SamplingParams, is_streaming: bool):
     if params.output_kind == target_type:
         return params
     elif is_streaming and params.output_kind == RequestOutputKind.FINAL_ONLY:
-        logger.warning("Request appears to be streaming, but got request type final only!")
+        logger.debug("Coercing FINAL_ONLY output to DELTA for streaming")
     elif not is_streaming and params.output_kind == RequestOutputKind.DELTA:
-        logger.warning("Request appears to not be streaming, but got request type delta!")
+        logger.debug("Coercing DELTA output to FINAL_ONLY for non-streaming")
 
     if not params.skip_clone:
         params = params.clone()

vllm_omni/model_executor/models/moss_tts/pipeline.py

@@ -1,8 +1,13 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 # Copyright 2026 OpenMOSS and the vLLM-Omni team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License").
 """Pipeline topology for all MOSS-TTS variants (2-stage: talker → codec)."""
 
+from vllm.sampling_params import RequestOutputKind
+
 from vllm_omni.config.stage_config import (
     PipelineConfig,
     StageExecutionType,
@@ -100,6 +105,10 @@ MOSS_TTS_LOCAL_PIPELINE = PipelineConfig(
             sampling_constraints={
                 "detokenize": False,
                 "stop_token_ids": [151645],
+                # The worker connector streams codes directly to the codec.
+                # This internal stage only needs to publish its terminal
+                # result; codec audio output remains incremental.
+                "output_kind": RequestOutputKind.FINAL_ONLY,
             },
         ),
         StagePipelineConfig(