[Perf][TTS] Add deadline-aware audio scheduling

Signed-off-by: Sy03 <1370724210@qq.com>
This commit is contained in:
Sy03 committed 2026-08-25 12:35:35 +08:00
1 parent 072bfc02dd
commit 5740ca8e01
8 files changed
+559 -19

No files matched your search

@@ -0,0 +1,39 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from types import SimpleNamespace
import pytest
import torch
from vllm_omni.core.sched.omni_generation_scheduler import OmniGenerationScheduler
pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
def test_generation_scheduler_records_audio_output_for_edf(mocker):
scheduler = OmniGenerationScheduler.__new__(OmniGenerationScheduler)
scheduler.chunk_transfer_adapter = SimpleNamespace(
playback_deadline_enabled=True,
record_audio_output=mocker.Mock(),
)
multimodal_output = {
"model_outputs": torch.zeros(24000),
"sr": torch.tensor(24000),
}
scheduler._record_playback_deadline_output("req", multimodal_output)
scheduler.chunk_transfer_adapter.record_audio_output.assert_called_once_with("req", 24000, 24000)
def test_generation_scheduler_skips_audio_accounting_when_edf_is_disabled(mocker):
scheduler = OmniGenerationScheduler.__new__(OmniGenerationScheduler)
scheduler.chunk_transfer_adapter = SimpleNamespace(
playback_deadline_enabled=False,
record_audio_output=mocker.Mock(),
)
scheduler._record_playback_deadline_output("req", {"model_outputs": torch.zeros(24000)})
scheduler.chunk_transfer_adapter.record_audio_output.assert_not_called()