vllm-omni/tests/e2e/online_serving/test_hunyuan_video_15_expansion.py

# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
"""
Tests of common diffusion feature combinations in online serving mode
for HunyuanVideo-1.5-T2V (480p).

Coverage:
- Default single-card deployment (H100, B200, or Ascend A2) — ``full_model``
- CPU offloading (1 H100 or Ascend A2) — ``core_model`` + ``advanced_model``
- CacheDiT + Layerwise CPU offloading (1 H100, B200, or Ascend A2) — ``full_model``
- CacheDiT + TP=2 + VAE patch parallel=2 (2 H100s, B200s, or Ascend A2s) — ``full_model``

HunyuanVideo-1.5 is a high-priority model. The CPU-offload row runs in the existing
CUDA PR (L2) and merge (L3) lanes. The shared default and CacheDiT / parallel rows
carry ``full_model`` (L4) marks, together with the video similarity suites in
``tests/e2e/accuracy/hunyuanvideo15_{t2v,i2v}/``.

From ``tests/``::

    pytest -s -v e2e/online_serving/test_hunyuan_video_15_expansion.py -m "core_model and diffusion" --run-level=core_model
    pytest -s -v e2e/online_serving/test_hunyuan_video_15_expansion.py -m "advanced_model and diffusion" --run-level=advanced_model
    pytest -s -v e2e/online_serving/test_hunyuan_video_15_expansion.py -m "full_model and diffusion" --run-level=full_model
"""

import pytest

from tests.helpers.mark import hardware_marks, hardware_test
from tests.helpers.runtime import OmniServer, OmniServerParams, OnlineOmniClient
from tests.helpers.stage_config import get_deploy_config_path

pytestmark = [pytest.mark.diffusion]

PROMPT = "A cat walking across a sunlit garden, cinematic lighting, slow motion."
NEGATIVE_PROMPT = "low quality, blurry, distorted"

MODEL = "hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-480p_t2v"
DEPLOY_CONFIG = get_deploy_config_path("hunyuan_video_15.yaml")

DEFAULT_TEST_PARAMS = [
    pytest.param(
        OmniServerParams(model=MODEL, stage_config_path=DEPLOY_CONFIG),
        id="default",
    )
]


def _get_diffusion_feature_cases(model: str):
    """Return diffusion feature cases for HunyuanVideo-1.5.

    Designed for up to two accelerators per issue #1832.
    Only the CPU-offload row is cheap enough for PR (L2) / merge (L3);
    CacheDiT / parallel combinations run nightly (L4).
    """
    return [
        # (1 accelerator) CPU offload
        pytest.param(
            OmniServerParams(
                model=model,
                server_args=[
                    "--enable-cpu-offload",
                ],
            ),
            id="single_card_cpu_offload",
            marks=[
                *hardware_marks(res={"cuda": "H100", "npu": "A2"}),
                pytest.mark.core_model,
                pytest.mark.advanced_model,
            ],
        ),
        # (1 accelerator) CacheDiT + Layerwise CPU offloading
        pytest.param(
            OmniServerParams(
                model=model,
                server_args=[
                    "--cache-backend",
                    "cache_dit",
                    "--enable-layerwise-offload",
                ],
            ),
            id="single_card_cachedit_layerwise",
            marks=[
                *hardware_marks(res={"cuda": ["H100", "B200"], "npu": "A2"}),
                pytest.mark.full_model,
            ],
        ),
        # (2 accelerators) CacheDiT + TP=2 + VAE patch parallel=2
        pytest.param(
            OmniServerParams(
                model=model,
                server_args=[
                    "--cache-backend",
                    "cache_dit",
                    "--tensor-parallel-size",
                    "2",
                    "--vae-patch-parallel-size",
                    "2",
                    "--vae-use-tiling",
                ],
            ),
            id="parallel_cachedit_tp2_vae2",
            marks=[
                *hardware_marks(res={"cuda": ["H100", "B200"], "npu": "A2"}, num_cards=2),
                pytest.mark.full_model,
            ],
        ),
    ]


def _run_hunyuan_video_15_t2v(
    omni_server: OmniServer,
    online_client: OnlineOmniClient,
) -> None:
    form_data = {
        "prompt": PROMPT,
        "negative_prompt": NEGATIVE_PROMPT,
        "height": 480,
        "width": 640,
        "num_frames": 5,
        "num_inference_steps": 2,
        "guidance_scale": 6.0,
        "seed": 42,
    }

    request_config = {
        "model": omni_server.model,
        "form_data": form_data,
    }

    online_client.send_video_diffusion_request(request_config)


@hardware_test(res={"cuda": ["H100", "B200"], "npu": "A2"}, num_cards=1)
@pytest.mark.full_model
@pytest.mark.parametrize("omni_server", DEFAULT_TEST_PARAMS, indirect=True)
def test_hunyuan_video_15_t2v_default(
    omni_server: OmniServer,
    online_client: OnlineOmniClient,
) -> None:
    """Generate a short video with the shared default GPU/NPU configuration."""
    _run_hunyuan_video_15_t2v(omni_server, online_client)


@pytest.mark.parametrize(
    "omni_server",
    _get_diffusion_feature_cases(MODEL),
    indirect=True,
)
def test_hunyuan_video_15_t2v_features(
    omni_server: OmniServer,
    online_client: OnlineOmniClient,
) -> None:
    """Exercise shared diffusion features for HunyuanVideo-1.5-T2V."""
    _run_hunyuan_video_15_t2v(omni_server, online_client)