[Test][Ascend] Share HunyuanVideo-1.5 E2E across GPU and NPU (#7547)

Signed-off-by: zengchuang <zengchuang3@huawei.com>

.buildkite/npu/test-npu-nightly.yml

@@ -235,6 +235,37 @@ steps:
       build.env("NIGHTLY") == "1" ||
       build.pull_request.labels includes "nightly-test"
     steps:
+      - label: ":full_moon: Diffusion X2V · HunyuanVideo-1.5 Function Test"
+        source_file_dependencies: diffusion_hunyuan_video_function
+        timeout_in_minutes: 180
+        mirror_hardwares: a2b3_npu_4
+        env:
+          VLLM_WORKER_MULTIPROC_METHOD: spawn
+          VLLM_ALLOW_LONG_MAX_MODEL_LEN: "1"
+          HF_TOKEN: "${HF_TOKEN}"
+          HF_HUB_OFFLINE: "1"
+        commands:
+          - pytest -s -v tests/e2e/online_serving/test_hunyuan_video_15_expansion.py -m "npu and A2" --run-level "full_model"
+      - label: ":full_moon: Diffusion X2V · HunyuanVideo-1.5 Perf Test"
+        source_file_dependencies: diffusion_hunyuan_video_perf
+        key: nightly-diffusion-x2v-performance-hunyuanvideo15
+        timeout_in_minutes: 180
+        mirror_hardwares: a3_npu_2
+        env:
+          VLLM_WORKER_MULTIPROC_METHOD: spawn
+          VLLM_ALLOW_LONG_MAX_MODEL_LEN: "1"
+          HF_TOKEN: "${HF_TOKEN}"
+          HF_HUB_OFFLINE: "1"
+          DIFFUSION_ATTENTION_BACKEND: TORCH_SDPA
+        commands:
+          - export DIFFUSION_BENCHMARK_DIR=tests/dfx/perf/results
+          - |
+            set +e
+            pytest -s -v tests/dfx/perf/scripts/run_diffusion_benchmark.py -m "npu and A3" --test-config-file tests/dfx/perf/tests/test_hunyuanvideo15_t2v_vllm_omni.json
+            EXIT=$$?
+            buildkite-agent artifact upload "tests/dfx/perf/results/diffusion_result_*.json"
+            buildkite-agent artifact upload "tests/dfx/perf/results/logs/*.log"
+            exit $$EXIT
       - label: ":full_moon: Diffusion X2V · Wan 2.2 Function Test"
         source_file_dependencies: diffusion_wan22_function
         mirror_hardwares: a2b3_npu_4

tests/dfx/perf/tests/test_hunyuanvideo15_t2v_vllm_omni.json

@@ -54,7 +54,8 @@
       {
         "hardware_marks": {
           "res": {
-            "cuda": ["H100", "B200"]
+            "cuda": ["H100", "B200"],
+            "npu": "A3"
           },
           "num_cards": 2
         }

tests/e2e/online_serving/test_hunyuan_video_15_expansion.py

@@ -4,14 +4,15 @@
 Tests of common diffusion feature combinations in online serving mode
 for HunyuanVideo-1.5-T2V (480p).
 
-Coverage (H100, since model cannot fit L4):
-- CPU offloading (1 GPU) — ``core_model`` + ``advanced_model``
-- CacheDiT + Layerwise CPU offloading (1 GPU) — ``full_model``
-- CacheDiT + TP=2 + VAE patch parallel=2 (2 GPUs) — ``full_model``
+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, so only the most basic single-card deployment
-row runs on every PR (L2) and on merge (L3). The heavyweight feature combinations stay
-nightly-only (L4), together with the video similarity suites in
+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/``::
@@ -23,8 +24,9 @@ From ``tests/``::
 
 import pytest
 
-from tests.helpers.mark import hardware_marks
+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]
 
@@ -32,17 +34,25 @@ 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 2x H100 environment per issue #1832.
+    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 GPU) CPU offload
+        # (1 accelerator) CPU offload
         pytest.param(
             OmniServerParams(
                 model=model,
@@ -52,12 +62,12 @@ def _get_diffusion_feature_cases(model: str):
             ),
             id="single_card_cpu_offload",
             marks=[
-                *hardware_marks(res={"cuda": "H100"}),
+                *hardware_marks(res={"cuda": "H100", "npu": "A2"}),
                 pytest.mark.core_model,
                 pytest.mark.advanced_model,
             ],
         ),
-        # (1 GPU) CacheDiT + Layerwise CPU offloading
+        # (1 accelerator) CacheDiT + Layerwise CPU offloading
         pytest.param(
             OmniServerParams(
                 model=model,
@@ -69,11 +79,11 @@ def _get_diffusion_feature_cases(model: str):
             ),
             id="single_card_cachedit_layerwise",
             marks=[
-                *hardware_marks(res={"cuda": ["H100", "B200"]}),
+                *hardware_marks(res={"cuda": ["H100", "B200"], "npu": "A2"}),
                 pytest.mark.full_model,
             ],
         ),
-        # (2 GPUs) CacheDiT + TP=2 + VAE patch parallel=2
+        # (2 accelerators) CacheDiT + TP=2 + VAE patch parallel=2
         pytest.param(
             OmniServerParams(
                 model=model,
@@ -89,23 +99,17 @@ def _get_diffusion_feature_cases(model: str):
             ),
             id="parallel_cachedit_tp2_vae2",
             marks=[
-                *hardware_marks(res={"cuda": ["H100", "B200"]}, num_cards=2),
+                *hardware_marks(res={"cuda": ["H100", "B200"], "npu": "A2"}, num_cards=2),
                 pytest.mark.full_model,
             ],
         ),
     ]
 
 
-@pytest.mark.parametrize(
-    "omni_server",
-    _get_diffusion_feature_cases(MODEL),
-    indirect=True,
-)
-def test_hunyuan_video_15_t2v(
+def _run_hunyuan_video_15_t2v(
     omni_server: OmniServer,
     online_client: OnlineOmniClient,
-):
-    """Diffusion feature coverage for HunyuanVideo-1.5-T2V on H100."""
+) -> None:
     form_data = {
         "prompt": PROMPT,
         "negative_prompt": NEGATIVE_PROMPT,
@@ -123,3 +127,27 @@ def test_hunyuan_video_15_t2v(
     }
 
     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)