[Test] Add example test cases for omni online (#1086)

Signed-off-by: wangyu31577 <wangyu31577@hundsun.com>
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Signed-off-by: yenuo26 <410167048@qq.com>
Co-authored-by: wangyu31577 <wangyu31577@hundsun.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

.buildkite/test-nightly.yaml

@@ -10,12 +10,13 @@ steps:
       queue: "cpu_queue_premerge"
 
   - label: "Omni Model Test with H100"
-    timeout_in_minutes: 180
+    timeout_in_minutes: 60
     depends_on: image-build
     if: build.env("NIGHTLY") == "1"
     commands:
       - export VLLM_WORKER_MULTIPROC_METHOD=spawn
       - pytest -s -v tests/e2e/online_serving/test_qwen3_omni_expansion.py
+      - pytest -s -v tests/examples/online_serving/test_qwen3_omni.py
     agents:
       queue: "mithril-h100-pool"
     plugins:
@@ -44,3 +45,24 @@ steps:
                 hostPath:
                   path: /mnt/hf-cache
                   type: DirectoryOrCreate
+
+
+
+  - label: "Omni Model Test"
+    timeout_in_minutes: 60
+    depends_on: image-build
+    commands:
+    - export VLLM_LOGGING_LEVEL=DEBUG
+    - export VLLM_WORKER_MULTIPROC_METHOD=spawn
+    - pytest -s -v tests/examples/online_serving/test_qwen2_5_omni.py
+    agents:
+      queue: "gpu_4_queue" # g6.12xlarge instance on AWS, has 4 L4 GPU
+    plugins:
+    - docker#v5.2.0:
+        image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
+        always-pull: true
+        propagate-environment: true
+        environment:
+        - "HF_HOME=/fsx/hf_cache"
+        volumes:
+        - "/fsx/hf_cache:/fsx/hf_cache"

tests/conftest.py

@@ -588,8 +588,6 @@ def convert_audio_to_text(audio_data):
     """
     Convert base64 encoded audio data to text using speech recognition.
     """
-    import whisper
-
     audio_data = base64.b64decode(audio_data)
     output_path = f"./test_{int(time.time())}"
     with open(output_path, "wb") as audio_file:
@@ -597,6 +595,13 @@ def convert_audio_to_text(audio_data):
 
     print(f"audio data is saved: {output_path}")
 
+    text = convert_audio_file_to_text(output_path=output_path)
+    return text
+
+
+def convert_audio_file_to_text(output_path):
+    import whisper
+
     model = whisper.load_model("base")
     text = model.transcribe(
         output_path,
@@ -604,6 +609,10 @@ def convert_audio_to_text(audio_data):
         word_timestamps=True,
         condition_on_previous_text=False,
     )["text"]
+    del model
+    if torch.cuda.is_available():
+        gc.collect()
+        torch.cuda.empty_cache()
     if text:
         return text
     else:
@@ -614,7 +623,6 @@ def merge_base64_and_convert_to_text(base64_list):
     """
     Merge a list of base64 encoded audio data and convert to text.
     """
-    import whisper
     from pydub import AudioSegment
 
     merged_audio = None
@@ -627,17 +635,8 @@ def merge_base64_and_convert_to_text(base64_list):
             merged_audio += seg
     output_path = f"./test_{int(time.time())}"
     merged_audio.export(output_path, format="wav")
-    model = whisper.load_model("base")
-    text = model.transcribe(
-        output_path,
-        temperature=0.0,
-        word_timestamps=True,
-        condition_on_previous_text=False,
-    )["text"]
-    if text:
-        return text
-    else:
-        return ""
+    text = convert_audio_file_to_text(output_path)
+    return text
 
 
 def modify_stage_config(
@@ -886,6 +885,7 @@ class OmniServer:
         model: str,
         serve_args: list[str],
         *,
+        port: int | None = None,
         env_dict: dict[str, str] | None = None,
     ) -> None:
         _run_pre_test_cleanup(enable_force=True)
@@ -896,7 +896,10 @@ class OmniServer:
         self.env_dict = env_dict
         self.proc: subprocess.Popen | None = None
         self.host = "127.0.0.1"
-        self.port = get_open_port()
+        if port is None:
+            self.port = get_open_port()
+        else:
+            self.port = port
 
     def _start_server(self) -> None:
         """Start the vLLM-Omni server subprocess."""

tests/e2e/stage_configs/qwen2_5_omni_ci.yaml

@@ -0,0 +1,106 @@
+# stage config for running qwen2.5-omni with architecture of OmniLLM.
+
+# The following config has been verified on 2x 24GB GPU (L4/RTX3090/RTX4090).
+# This config is optimized for CI e2e tests.
+stage_args:
+  - stage_id: 0
+    runtime:
+      process: true            # Run this stage in a separate process
+      devices: "0"            # Visible devices for this stage (CUDA_VISIBLE_DEVICES/torch.cuda.set_device)
+      max_batch_size: 1
+    engine_args:
+      model_stage: thinker
+      model_arch: Qwen2_5OmniForConditionalGeneration
+      worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker
+      scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
+      max_model_len: 32768
+      max_num_batched_tokens: 32768
+      max_num_seqs: 1
+      gpu_memory_utilization: 0.8
+      skip_mm_profiling: true
+      enforce_eager: true  # Now we only support eager mode
+      trust_remote_code: true
+      engine_output_type: latent
+      enable_prefix_caching: false
+    is_comprehension: true
+    final_output: true
+    final_output_type: text
+    default_sampling_params:
+      temperature: 0.0
+      top_p: 1.0
+      top_k: -1
+      max_tokens: 128
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.1
+  - stage_id: 1
+    runtime:
+      process: true
+      devices: "1"
+      max_batch_size: 1
+    engine_args:
+      model_stage: talker
+      model_arch: Qwen2_5OmniForConditionalGeneration
+      worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker
+      scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
+      max_model_len: 32768
+      max_num_batched_tokens: 32768
+      max_num_seqs: 1
+      gpu_memory_utilization: 0.8
+      skip_mm_profiling: true
+      enforce_eager: true
+      trust_remote_code: true
+      enable_prefix_caching: false
+      engine_output_type: latent
+    engine_input_source: [0]
+    custom_process_input_func: vllm_omni.model_executor.stage_input_processors.qwen2_5_omni.thinker2talker
+    default_sampling_params:
+      temperature: 0.9
+      top_p: 0.8
+      top_k: 40
+      max_tokens: 128
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.05
+      stop_token_ids: [8294]
+  - stage_id: 2
+    runtime:
+      process: true
+      devices: "0"            # Example: use a different GPU than the previous stage; use "0" if single GPU
+      max_batch_size: 1
+    engine_args:
+      model_stage: code2wav
+      model_arch: Qwen2_5OmniForConditionalGeneration
+      worker_cls: vllm_omni.worker.gpu_generation_worker.GPUGenerationWorker
+      scheduler_cls: vllm_omni.core.sched.omni_generation_scheduler.OmniGenerationScheduler
+      gpu_memory_utilization: 0.15
+      enforce_eager: true
+      trust_remote_code: true
+      enable_prefix_caching: false
+      engine_output_type: audio
+      max_num_batched_tokens: 4069
+    engine_input_source: [1]
+    final_output: true
+    final_output_type: audio
+    default_sampling_params:
+      temperature: 0.0
+      top_p: 1.0
+      top_k: -1
+      max_tokens: 128
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.1
+
+# Top-level runtime config (concise): default windows and stage edges
+runtime:
+  enabled: true
+  defaults:
+    window_size: -1             # Simplified: trigger downstream only after full upstream completion
+    max_inflight: 1             # Simplified: process serially within each stage
+  edges:
+    - from: 0                   # thinker → talker: trigger only after receiving full input (-1)
+      to: 1
+      window_size: -1
+    - from: 1                   # talker → code2wav: trigger only after receiving full input (-1)
+      to: 2
+      window_size: -1

tests/examples/online_serving/test_qwen2_5_omni.py

@@ -0,0 +1,271 @@
+"""
+Example online tests for Qwen2.5-Omni-7B model.
+"""
+
+import os
+
+os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
+
+import re
+import subprocess
+from pathlib import Path
+
+import pytest
+
+from tests.conftest import OmniServer, convert_audio_file_to_text, cosine_similarity_text
+
+models = ["Qwen/Qwen2.5-Omni-7B"]
+
+
+stage_configs = [str(Path(__file__).parent.parent.parent / "e2e" / "stage_configs" / "qwen2_5_omni_ci.yaml")]
+
+example_dir = str(Path(__file__).parent.parent.parent.parent / "examples" / "online_serving" / "qwen2_5_omni")
+# Create parameter combinations for model and stage config
+test_params = [(model, stage_config) for model in models for stage_config in stage_configs]
+
+
+@pytest.fixture(scope="module")
+def omni_server(request):
+    """Start vLLM-Omni server as a subprocess with actual model weights.
+    Uses module scope so the server starts only once per test module.
+    Multi-stage initialization can take 10-20+ minutes.
+    """
+    model, stage_config_path = request.param
+
+    print(f"Starting OmniServer with model: {model}")
+    print("This may take 10-20+ minutes for initialization...")
+
+    with OmniServer(
+        model, ["--stage-configs-path", stage_config_path, "--stage-init-timeout", "120"], port=8091
+    ) as server:
+        print("OmniServer started successfully")
+        yield server
+        print("OmniServer stopped")
+
+
+def run_cmd(command):
+    result = subprocess.run(
+        command,
+        capture_output=True,
+        text=True,
+    )
+
+    if result.returncode != 0:
+        print(f"STDERR: {result.stderr}")
+        raise subprocess.CalledProcessError(result.returncode, command)
+
+    all_output = result.stdout
+    print(f"All output:\n{all_output}")
+    return all_output
+
+
+def extract_content_after_keyword(keywords, text):
+    matches = re.findall(rf"{keywords}\s*(.+)", text, re.DOTALL)
+
+    if not matches:
+        raise AssertionError(f"Keywords {keywords} not found in provided text output")
+    return matches[0]
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_send_multimodal_request_001(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "mixed_modalities",
+    ]
+
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("Chat completion output from text:", result)
+
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"text content is: {text_content}")
+    print(f"audio content is: {audio_content}")
+
+    assert all(keyword in text_content for keyword in ["baby", "book"]), (
+        "The output does not contain any of the keywords in video description."
+    )
+    # There is currently an issue with incorrect image descriptions.
+    # assert "cherry blossom" in text_content, "The output does not contain any of the keywords in image description."
+    assert "lamb" in text_content, "The output does not contain any of the keywords in audio description."
+
+    similarity = cosine_similarity_text(audio_content.lower(), text_content.lower())
+    print(f"similarity is: {similarity}")
+    assert similarity > 0.9, "The audio content is not same as the text"
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_send_multimodal_request_002(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "mixed_modalities",
+        "--prompt",
+        "Analyze all the media content and provide a comprehensive summary.",
+    ]
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("Chat completion output from text:", result)
+
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"text content is: {text_content}")
+    assert all(keyword in text_content for keyword in ["baby", "book"]), (
+        "The output does not contain any of the keywords in video description."
+    )
+    # There is currently an issue with incorrect image descriptions.
+    # assert "cherry blossom" in text_content, "The output does not contain any of the keywords in image description."
+    assert "lamb" in text_content, "The output does not contain any of the keywords in audio description."
+
+    print(f"audio content is: {audio_content}")
+    similarity = cosine_similarity_text(audio_content.lower(), text_content.lower())
+    print(f"similarity is: {similarity}")
+    assert similarity > 0.9, "The audio content is not same as the text"
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_send_multimodal_request_003(omni_server) -> None:
+    command = ["bash", os.path.join(example_dir, "run_curl_multimodal_generation.sh"), "mixed_modalities"]
+
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("Output of request:", result)
+
+    # Verify text output same as audio output
+    print(f"text content is: {text_content}")
+    assert all(keyword in text_content for keyword in ["baby", "book"]), (
+        "The output does not contain any of the keywords in video description."
+    )
+    # There is currently an issue with incorrect image descriptions.
+    # assert "cherry blossom" in text_content, "The output does not contain any of the keywords in image description."
+    assert "lamb" in text_content, "The output does not contain any of the keywords in audio description."
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_modality_control_001(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "mixed_modalities",
+        "--modalities",
+        "text",
+    ]
+
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("Chat completion output from text:", result)
+
+    # Verify text output
+    print(f"text content is: {text_content}")
+    assert all(keyword in text_content for keyword in ["baby", "book"]), (
+        "The output does not contain any of the keywords in video description."
+    )
+    # There is currently an issue with incorrect image descriptions.
+    # assert "cherry blossom" in text_content, "The output does not contain any of the keywords in image description."
+    assert "lamb" in text_content, "The output does not contain any of the keywords in audio description."
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_modality_control_002(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "mixed_modalities",
+        "--modalities",
+        "audio",
+    ]
+
+    run_cmd(command)
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"audio content is: {audio_content}")
+    assert all(keyword in audio_content for keyword in ["baby", "book"]), (
+        "The output does not contain any of the keywords in video description."
+    )
+    # There is currently an issue with incorrect image descriptions.
+    # assert "cherry blossom" in audio_content, "The output does not contain any of the keywords in image description."
+    assert "lamb" in audio_content, "The output does not contain any of the keywords in audio description."
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_modality_control_003(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "mixed_modalities",
+        "--modalities",
+        "audio,text",
+    ]
+
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("Chat completion output from text:", result)
+
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"text content is: {text_content}")
+    assert all(keyword in text_content for keyword in ["baby", "book"]), (
+        "The output does not contain any of the keywords in video description."
+    )
+    # There is currently an issue with incorrect image descriptions.
+    # assert "cherry blossom" in text_content, "The output does not contain any of the keywords in image description."
+    assert "lamb" in text_content, "The output does not contain any of the keywords in audio description."
+
+    print(f"audio content is: {audio_content}")
+    similarity = cosine_similarity_text(audio_content.lower(), text_content.lower())
+    print(f"similarity is: {similarity}")
+    assert similarity > 0.9, "The audio content is not same as the text"
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.skip(reason="There is a known issue with stream error.")
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_stream_001(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "mixed_modalities",
+        "--stream",
+    ]
+
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("content:", result)
+
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"text content is: {text_content}")
+    assert all(keyword in text_content for keyword in ["baby", "book"]), (
+        "The output does not contain any of the keywords in video description."
+    )
+    # There is currently an issue with incorrect image descriptions.
+    # assert "cherry blossom" in text_content, "The output does not contain any of the keywords in image description."
+    assert "lamb" in text_content, "The output does not contain any of the keywords in audio description."
+
+    print(f"audio content is: {audio_content}")
+    similarity = cosine_similarity_text(audio_content.lower(), text_content.lower())
+    print(f"similarity is: {similarity}")
+    assert similarity > 0.9, "The audio content is not same as the text"
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+# TODO: test local web ui

tests/examples/online_serving/test_qwen3_omni.py

@@ -0,0 +1,230 @@
+"""
+Example Online tests for Qwen3-Omni model.
+"""
+
+import os
+
+os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
+
+import re
+import subprocess
+from pathlib import Path
+
+import pytest
+
+from tests.conftest import OmniServer, convert_audio_file_to_text, cosine_similarity_text
+
+models = ["Qwen/Qwen3-Omni-30B-A3B-Instruct"]
+
+
+stage_configs = [str(Path(__file__).parent.parent.parent / "e2e" / "stage_configs" / "qwen3_omni_ci.yaml")]
+
+example_dir = str(Path(__file__).parent.parent.parent.parent / "examples" / "online_serving" / "qwen3_omni")
+# Create parameter combinations for model and stage config
+test_params = [(model, stage_config) for model in models for stage_config in stage_configs]
+
+
+@pytest.fixture(scope="module")
+def omni_server(request):
+    """Start vLLM-Omni server as a subprocess with actual model weights.
+    Uses module scope so the server starts only once per test module..
+    Multi-stage initialization can take 10-20+ minutes.
+    """
+    model, stage_config_path = request.param
+
+    print(f"Starting OmniServer with model: {model}")
+    print("This may take 10-20+ minutes for initialization...")
+
+    with OmniServer(
+        model, ["--stage-configs-path", stage_config_path, "--stage-init-timeout", "120"], port=8091
+    ) as server:
+        print("OmniServer started successfully")
+        yield server
+        print("OmniServer stopped")
+
+
+def run_cmd(command):
+    result = subprocess.run(
+        command,
+        capture_output=True,
+        text=True,
+    )
+
+    if result.returncode != 0:
+        print(f"STDERR: {result.stderr}")
+        raise subprocess.CalledProcessError(result.returncode, command)
+
+    all_output = result.stdout
+    print(f"All output:\n{all_output}")
+    return all_output
+
+
+def extract_content_after_keyword(keywords, text):
+    matches = re.findall(rf"{keywords}\s*(.+)", text, re.DOTALL)
+
+    if not matches:
+        raise AssertionError(f"Keywords {keywords} not found in provided text output")
+    return matches[0]
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_send_multimodal_request_001(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "use_image",
+    ]
+
+    result = run_cmd(command)
+    text_content = extract_content_after_keyword("Chat completion output from text:", result)
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"text content is: {text_content}")
+    print(f"audio content is: {audio_content}")
+
+    assert "cherry blossom" in text_content, "The output does not contain any of the keywords."
+
+    similarity = cosine_similarity_text(audio_content.lower(), text_content.lower())
+    print(f"similarity is: {similarity}")
+    assert similarity > 0.9, "The audio content is not same as the text"
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_send_multimodal_request_002(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "use_video",
+        "--prompt",
+        "What are the main activities shown in this video?",
+    ]
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("Chat completion output from text:", result)
+
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"text content is: {text_content}")
+    print(f"audio content is: {audio_content}")
+    assert all(keyword in text_content for keyword in ["baby", "book"]), (
+        "The output does not contain any of the keywords."
+    )
+    similarity = cosine_similarity_text(audio_content.lower(), text_content.lower())
+    print(f"similarity is: {similarity}")
+    assert similarity > 0.9, "The audio content is not same as the text"
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_send_multimodal_request_003(omni_server) -> None:
+    command = ["bash", os.path.join(example_dir, "run_curl_multimodal_generation.sh"), "use_image"]
+
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("Output of request:", result)
+
+    # Verify text output same as audio output
+    print(f"text content is: {text_content}")
+    assert "cherry blossom" in text_content, "The output does not contain any of the keywords."
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_modality_control_001(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "use_image",
+        "--modalities",
+        "text",
+    ]
+
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("Chat completion output from text:", result)
+
+    # Verify text output
+    print(f"text content is: {text_content}")
+    assert "cherry blossom" in text_content, "The output does not contain any of the keywords."
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_modality_control_002(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "use_image",
+        "--modalities",
+        "audio",
+    ]
+
+    run_cmd(command)
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"audio content is: {audio_content}")
+    assert "cherry blossom" in audio_content, "The output does not contain any of the keywords."
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_modality_control_003(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "use_image",
+        "--modalities",
+        "audio,text",
+    ]
+
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("Chat completion output from text:", result)
+
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"text content is: {text_content}")
+    assert "cherry blossom" in audio_content, "The output does not contain any of the keywords."
+    print(f"audio content is: {audio_content}")
+    similarity = cosine_similarity_text(audio_content.lower(), text_content.lower())
+    print(f"similarity is: {similarity}")
+    assert similarity > 0.9, "The audio content is not same as the text"
+
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+def test_stream_001(omni_server) -> None:
+    command = [
+        "python",
+        os.path.join(example_dir, "openai_chat_completion_client_for_multimodal_generation.py"),
+        "--query-type",
+        "use_image",
+        "--stream",
+    ]
+
+    result = run_cmd(command)
+
+    text_content = extract_content_after_keyword("content:", result)
+
+    # Verify text output same as audio output
+    audio_content = convert_audio_file_to_text(output_path="./audio_0.wav")
+    print(f"text content is: {text_content}")
+    assert "cherry blossom" in audio_content, "The output does not contain any of the keywords."
+    print(f"audio content is: {audio_content}")
+    similarity = cosine_similarity_text(audio_content.lower(), text_content.lower())
+    print(f"similarity is: {similarity}")
+    assert similarity > 0.9, "The audio content is not same as the text"
+    # TODO: Verify the E2E latency after confirmation baseline.
+
+
+# TODO: test local web ui