[Bugfix] Use Seed-TTS reference audio with MiniCPM-o 4.5 (#6628)

Signed-off-by: akshatvishu <akshatnayak197@gmail.com>
This commit is contained in:
akshatvishu authored and GitHub committed 2026-08-31 10:27:34 +08:00
1 parent 4cca28e9bd
commit f321cdbe16
10 files changed
+529 -56

No files matched your search

@@ -1,12 +1,20 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
"""E2E online tests for MiniCPM-o 4.5 multimodal input and audio/text output.
Exercises async chunk streaming (``--async-chunk``) across separate Thinker,
Talker, and Code2Wav stages.
"""
import json
import os
import tempfile
from copy import deepcopy
from pathlib import Path
import pytest
from vllm.logger import DEFAULT_LOGGING_CONFIG
from tests.helpers.mark import hardware_test
from tests.helpers.media import generate_synthetic_audio, generate_synthetic_image, generate_synthetic_video
@@ -23,6 +31,23 @@ _CI_DEPLOY = modify_stage_config(
},
)
_PROMPT_LOG_TEMP_DIR = tempfile.TemporaryDirectory(prefix="minicpmo45-prompt-selection-")
_PROMPT_SELECTION_LOG = Path(_PROMPT_LOG_TEMP_DIR.name) / "prompt-selection.log"
_PROMPT_LOGGING_CONFIG_PATH = Path(_PROMPT_LOG_TEMP_DIR.name) / "logging.json"
_PROMPT_LOGGING_CONFIG = deepcopy(DEFAULT_LOGGING_CONFIG)
_PROMPT_LOGGING_CONFIG["handlers"]["minicpmo_prompt_selection"] = {
"class": "logging.FileHandler",
"filename": str(_PROMPT_SELECTION_LOG),
"formatter": "vllm",
"level": "DEBUG",
}
_PROMPT_LOGGING_CONFIG["loggers"]["vllm_omni.model_executor.models.minicpmo_4_5.minicpmo_4_5_code2wav"] = {
"handlers": ["minicpmo_prompt_selection"],
"level": "DEBUG",
"propagate": True,
}
_PROMPT_LOGGING_CONFIG_PATH.write_text(json.dumps(_PROMPT_LOGGING_CONFIG), encoding="utf-8")
test_params = [
pytest.param(
OmniServerParams(
@@ -30,6 +55,10 @@ test_params = [
stage_config_path=_CI_DEPLOY,
use_stage_cli=False,
server_args=["--trust-remote-code", "--async-chunk"],
env_dict={
"VLLM_CONFIGURE_LOGGING": "1",
"VLLM_LOGGING_CONFIG_PATH": str(_PROMPT_LOGGING_CONFIG_PATH),
},
),
id="async_chunk",
),
@@ -74,6 +103,18 @@ _TTS_EXTRA_BODY = {
}
def _prompt_selection_log_offset() -> int:
return _PROMPT_SELECTION_LOG.stat().st_size if _PROMPT_SELECTION_LOG.exists() else 0
def _read_prompt_selection_log(offset: int) -> str:
if not _PROMPT_SELECTION_LOG.exists():
return ""
with open(_PROMPT_SELECTION_LOG, "rb") as log_file:
log_file.seek(offset)
return log_file.read().decode("utf-8")
@pytest.mark.core_model
@pytest.mark.advanced_model
@pytest.mark.omni
@@ -126,6 +167,51 @@ def test_text_to_audio_001(omni_server, openai_client) -> None:
openai_client.send_omni_request(request_config)
@pytest.mark.advanced_model
@pytest.mark.omni
@hardware_test(res={"cuda": "H100", "npu": "A3"}, num_cards=1)
@pytest.mark.parametrize("omni_server", test_params, indirect=True)
def test_text_to_audio_with_reference_audio(omni_server, openai_client) -> None:
request_config = {
"model": omni_server.model,
"messages": dummy_messages_from_mix_data(system_prompt=get_system_prompt(), content_text=get_prompt()),
"stream": True,
"modalities": ["text", "audio"],
"extra_body": {
**_TTS_EXTRA_BODY,
"ref_audio": f"data:audio/wav;base64,{generate_synthetic_audio(5, 1)['base64']}",
},
}
log_offset = _prompt_selection_log_offset()
responses = openai_client.send_omni_request(request_config)
prompt_log = _read_prompt_selection_log(log_offset)
assert responses[0].success
assert responses[0].audio_bytes
assert "prompt_cache_id=runtime-ref-" in prompt_log
assert "minicpmo45_ref_" in prompt_log
@pytest.mark.advanced_model
@pytest.mark.omni
@hardware_test(res={"cuda": "H100", "npu": "A3"}, num_cards=1)
@pytest.mark.parametrize("omni_server", test_params, indirect=True)
def test_text_to_audio_with_default_reference(omni_server, openai_client) -> None:
request_config = {
"model": omni_server.model,
"messages": dummy_messages_from_mix_data(system_prompt=get_system_prompt(), content_text=get_prompt()),
"stream": True,
"modalities": ["text", "audio"],
"extra_body": _TTS_EXTRA_BODY,
}
responses = openai_client.send_omni_request(request_config)
assert responses[0].success
assert responses[0].audio_bytes
@pytest.mark.full_model
@pytest.mark.omni
@hardware_test(res={"cuda": "H100", "npu": "A3"}, num_cards=1)