vllm-omni/tests/e2e/online_serving/test_qwen3_tts_base_expansion.py

# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
"""
E2E Online tests for Qwen3-TTS model with text input and audio output.

These tests verify the /v1/audio/speech endpoint works correctly with
actual model inference, not mocks.
"""

import os

import pytest

from tests.helpers.mark import hardware_test
from tests.helpers.media import get_asset_path
from tests.helpers.runtime import OmniServerParams
from tests.helpers.stage_config import get_deploy_config_path

pytestmark = [pytest.mark.full_model, pytest.mark.tts]

os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"

MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"

# See tests/e2e/online_serving/test_qwen3_tts_base.py for the vendored-asset rationale.
REF_AUDIO_URL = get_asset_path("qwen3_tts/clone_2.wav", as_data_url=True)
REF_TEXT = "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it! And thanks to you."


def get_prompt(prompt_type="text"):
    """Text prompt for text-to-audio tests (same as test_qwen3_omni - beijing test case)."""
    prompts = {
        "text": "The weather is nice today, perfect for a walk in the park.",
    }
    return prompts.get(prompt_type, prompts["text"])


def get_max_batch_size(size_type="few"):
    """Batch size for concurrent requests (same as test_qwen3_omni)."""
    batch_sizes = {"few": 5, "medium": 100, "large": 256}
    return batch_sizes.get(size_type, 5)


tts_async_chunk_server_params = [
    pytest.param(
        OmniServerParams(
            model=MODEL,
            stage_config_path=get_deploy_config_path("qwen3_tts.yaml"),
            server_args=["--trust-remote-code"],
        ),
        id="async_chunk",
    ),
]

# Synchronous (no async-chunk) variant — ``--no-async-chunk`` alone
# flips the deploy yaml's bool and the pipeline dispatches to the
# end-to-end codec processor. No variant yaml / pipeline needed.
tts_no_async_chunk_server_params = [
    pytest.param(
        OmniServerParams(
            model=MODEL,
            stage_config_path=get_deploy_config_path("qwen3_tts.yaml"),
            server_args=["--trust-remote-code", "--no-async-chunk"],
        ),
        id="no_async_chunk",
    ),
]

DEFAULT_AUDIO_SPEECH_TIMEOUT_S = 180.0


@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=1)
@pytest.mark.parametrize(
    "omni_server",
    tts_async_chunk_server_params + tts_no_async_chunk_server_params,
    indirect=True,
)
def test_voice_clone_streaming_001(omni_server, online_client) -> None:
    """
    Test text input processing and audio output via OpenAI API.
    Deploy Setting: default yaml
    Input Modal: text
    Output Modal: audio
    Input Setting: stream=True
    Datasets: few requests
    """

    request_config = {
        "model": omni_server.model,
        "input": get_prompt(),
        "stream": True,
        "stream_format": "audio",
        "response_format": "wav",
        "task_type": "Base",
        "voice": "clone",
        "ref_audio": REF_AUDIO_URL,
        "ref_text": REF_TEXT,
    }
    online_client.send_audio_speech_request(request_config, request_num=get_max_batch_size("few"))


@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=1)
@pytest.mark.parametrize(
    "omni_server",
    tts_async_chunk_server_params + tts_no_async_chunk_server_params,
    indirect=True,
)
def test_response_format_001(omni_server, online_client) -> None:
    """
    Test text input processing and audio output via OpenAI API.
    Deploy Setting: default yaml
    Input Modal: text
    Output Modal: audio
    Input Setting: non-stream PCM
    Datasets: few requests
    """

    request_config = {
        "model": omni_server.model,
        "input": get_prompt(),
        "response_format": "pcm",
        "task_type": "Base",
        "voice": "clone",
        "ref_audio": REF_AUDIO_URL,
        "ref_text": REF_TEXT,
    }
    online_client.send_audio_speech_request(request_config)


@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=1)
@pytest.mark.parametrize("omni_server", tts_async_chunk_server_params, indirect=True)
def test_xvector_then_icl_same_ref_audio_keeps_engine_alive(omni_server, online_client) -> None:
    """Regression for #5049: x-vector-only then ICL on the same ``ref_audio``.

    Pre-fix, readiness was mode-agnostic: an x-vector request marked the
    artifact ready (embedding only, no ``ref_code``), so the following ICL
    request took the artifact-only path and killed EngineCore. After the fix
    (readiness keyed by ``(artifact_key, x_vector_only)``), both modes succeed
    and the server stays up. Only runs on the ``async_chunk`` server param.
    """
    base_request = {
        "model": omni_server.model,
        "input": get_prompt(),
        "stream": False,
        "timeout": DEFAULT_AUDIO_SPEECH_TIMEOUT_S,
        "response_format": "wav",
        "task_type": "Base",
        "ref_audio": REF_AUDIO_URL,
        "min_audio_bytes": 1,
    }

    online_client.send_audio_speech_request({**base_request, "x_vector_only_mode": True})
    online_client.send_audio_speech_request({**base_request, "x_vector_only_mode": False, "ref_text": REF_TEXT})
    online_client.send_audio_speech_request({**base_request, "x_vector_only_mode": True})