[CI/Build] Share nightly CUDA YAML between H100/L4 and B200 (#7028)
Signed-off-by: wangyu <410167048@qq.com> Signed-off-by: [Your Name] [Your Email] Signed-off-by: [Your Name] <your.email@example.com> Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: amy-why-3459 <wuhaiyan17@huawei.com>
This commit is contained in:
78 files changed
+437
-321
No files matched your search
@@ -33,19 +33,13 @@ NEGATIVE_PROMPT = "low quality, blurry, distorted"
|
||||
|
||||
MODEL = "hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-480p_t2v"
|
||||
|
||||
SINGLE_CARD_MARKS = hardware_marks(res={"cuda": "H100"})
|
||||
PARALLEL_MARKS = hardware_marks(res={"cuda": "H100"}, num_cards=2)
|
||||
|
||||
# Only the most basic single-card deployment row is cheap enough for PR (L2) / merge (L3)
|
||||
# CI; the CacheDiT / parallel feature combinations run nightly (L4).
|
||||
_CORE_MARKS = [pytest.mark.core_model, pytest.mark.advanced_model]
|
||||
_FULL_MARKS = [pytest.mark.full_model]
|
||||
|
||||
|
||||
def _get_diffusion_feature_cases(model: str):
|
||||
"""Return diffusion feature cases for HunyuanVideo-1.5.
|
||||
|
||||
Designed for 2x H100 environment 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
|
||||
@@ -57,7 +51,11 @@ def _get_diffusion_feature_cases(model: str):
|
||||
],
|
||||
),
|
||||
id="single_card_cpu_offload",
|
||||
marks=SINGLE_CARD_MARKS + _CORE_MARKS,
|
||||
marks=[
|
||||
*hardware_marks(res={"cuda": "H100"}),
|
||||
pytest.mark.core_model,
|
||||
pytest.mark.advanced_model,
|
||||
],
|
||||
),
|
||||
# (1 GPU) CacheDiT + Layerwise CPU offloading
|
||||
pytest.param(
|
||||
@@ -70,7 +68,10 @@ def _get_diffusion_feature_cases(model: str):
|
||||
],
|
||||
),
|
||||
id="single_card_cachedit_layerwise",
|
||||
marks=SINGLE_CARD_MARKS + _FULL_MARKS,
|
||||
marks=[
|
||||
*hardware_marks(res={"cuda": ["H100", "B200"]}),
|
||||
pytest.mark.full_model,
|
||||
],
|
||||
),
|
||||
# (2 GPUs) CacheDiT + TP=2 + VAE patch parallel=2
|
||||
pytest.param(
|
||||
@@ -87,7 +88,10 @@ def _get_diffusion_feature_cases(model: str):
|
||||
],
|
||||
),
|
||||
id="parallel_cachedit_tp2_vae2",
|
||||
marks=PARALLEL_MARKS + _FULL_MARKS,
|
||||
marks=[
|
||||
*hardware_marks(res={"cuda": ["H100", "B200"]}, num_cards=2),
|
||||
pytest.mark.full_model,
|
||||
],
|
||||
),
|
||||
]
|
||||
|
||||
|
||||
@@ -20,7 +20,7 @@ PROMPT = "a robotic arm picks up a red block"
|
||||
NEGATIVE_PROMPT = "low quality, blurry, watermark, text"
|
||||
DEFAULT_SAMPLING_PARAMS = '{"0":{"num_frames":81,"num_inference_steps":40,"guidance_scale":6.0}}'
|
||||
|
||||
SINGLE_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": "H100"})
|
||||
SINGLE_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": ["H100", "B200"]})
|
||||
|
||||
|
||||
def _get_diffusion_feature_cases(model: str):
|
||||
|
||||
@@ -1,5 +1,5 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
|
||||
|
||||
"""
|
||||
L4 expansion coverage for ``robbyant/lingbot-video-dense-1.3b``.
|
||||
@@ -26,7 +26,7 @@ MODEL = "robbyant/lingbot-video-dense-1.3b"
|
||||
PROMPT = "a robotic arm picks up a red block"
|
||||
NEGATIVE_PROMPT = "low quality, blurry, watermark, text"
|
||||
|
||||
SINGLE_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": "H100"})
|
||||
SINGLE_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": ["H100", "B200"]})
|
||||
|
||||
|
||||
def _get_diffusion_feature_cases(model: str):
|
||||
|
||||
@@ -20,7 +20,7 @@ PROMPT = "a robotic arm picks up a red block"
|
||||
NEGATIVE_PROMPT = "low quality, blurry, watermark, text"
|
||||
SMOKE_DEFAULT_SAMPLING_PARAMS = '{"0":{"num_frames":81,"num_inference_steps":2,"guidance_scale":3.0}}'
|
||||
|
||||
SINGLE_CARD_MARKS = hardware_marks(res={"cuda": "H100"})
|
||||
SINGLE_CARD_MARKS = hardware_marks(res={"cuda": ["H100", "B200"]})
|
||||
|
||||
|
||||
def _get_server_cases(model: str):
|
||||
|
||||
@@ -51,7 +51,7 @@ def _cached_server_vad_artifact() -> None:
|
||||
)
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "H100", "npu": "A3"}, num_cards=1)
|
||||
@hardware_test(res={"cuda": ["H100", "B200"], "npu": "A3"}, num_cards=1)
|
||||
@pytest.mark.parametrize("omni_server", SERVER_PARAMS, indirect=True)
|
||||
def test_duplex_admission_and_expiry_reaper(omni_server, tmp_path: Path) -> None:
|
||||
args = multi_session_args(
|
||||
@@ -79,7 +79,7 @@ def test_duplex_admission_and_expiry_reaper(omni_server, tmp_path: Path) -> None
|
||||
# still talking, which only holds when the duplex pipeline sustains real-time
|
||||
# throughput. The current NPU stack runs several times slower than real time,
|
||||
# so it never reaches a mid-stream decision point.
|
||||
@hardware_test(res={"cuda": "H100"}, num_cards=1)
|
||||
@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards=1)
|
||||
@pytest.mark.parametrize("omni_server", SERVER_PARAMS, indirect=True)
|
||||
def test_duplex_soft_interrupt(omni_server, tmp_path: Path) -> None:
|
||||
input_wav = validated_soft_interrupt_wav()
|
||||
@@ -112,7 +112,7 @@ def test_duplex_soft_interrupt(omni_server, tmp_path: Path) -> None:
|
||||
assert result["followup_response_transcript_ok"] is True
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "H100"}, num_cards=1)
|
||||
@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards=1)
|
||||
@pytest.mark.parametrize("omni_server", SERVER_PARAMS, indirect=True)
|
||||
def test_duplex_server_vad_hard_interrupt(_cached_server_vad_artifact, omni_server) -> None:
|
||||
result = asyncio.run(
|
||||
|
||||
@@ -1,5 +1,5 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
|
||||
|
||||
"""E2E online serving test for π0 (Pi-Zero) OpenPI websocket serving.
|
||||
|
||||
@@ -58,7 +58,7 @@ test_params = [
|
||||
|
||||
@pytest.mark.full_model
|
||||
@pytest.mark.diffusion
|
||||
@hardware_test(res={"cuda": "H100"})
|
||||
@hardware_test(res={"cuda": ["H100", "B200"]})
|
||||
@pytest.mark.parametrize("omni_server", test_params, indirect=True)
|
||||
def test_pi0_openpi_online(omni_server):
|
||||
try:
|
||||
|
||||
@@ -91,7 +91,7 @@ def _assert_audio_outputs(responses: list[OmniResponse], *, expect_text: bool) -
|
||||
|
||||
@pytest.mark.full_model
|
||||
@pytest.mark.omni
|
||||
@hardware_test(res={"cuda": "H100"}, num_cards=4)
|
||||
@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards=4)
|
||||
@pytest.mark.parametrize("omni_server", test_params, indirect=True)
|
||||
def test_text_only_batch_uses_multi_replica_talker(omni_server, online_client) -> None:
|
||||
request_config = {
|
||||
@@ -108,7 +108,7 @@ def test_text_only_batch_uses_multi_replica_talker(omni_server, online_client) -
|
||||
|
||||
@pytest.mark.full_model
|
||||
@pytest.mark.omni
|
||||
@hardware_test(res={"cuda": "H100"}, num_cards=4)
|
||||
@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards=4)
|
||||
@pytest.mark.parametrize("omni_server", test_params, indirect=True)
|
||||
def test_text_to_audio_stream_batch_uses_multi_replica_vocoder(omni_server, online_client) -> None:
|
||||
request_config = {
|
||||
@@ -125,7 +125,7 @@ def test_text_to_audio_stream_batch_uses_multi_replica_vocoder(omni_server, onli
|
||||
|
||||
@pytest.mark.full_model
|
||||
@pytest.mark.omni
|
||||
@hardware_test(res={"cuda": "H100"}, num_cards=4)
|
||||
@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards=4)
|
||||
@pytest.mark.parametrize("omni_server", test_params, indirect=True)
|
||||
def test_mixed_modal_stream_batch_generates_text_and_audio(omni_server, online_client) -> None:
|
||||
request_config = {
|
||||
|
||||
@@ -69,7 +69,7 @@ tts_no_async_chunk_server_params = [
|
||||
DEFAULT_AUDIO_SPEECH_TIMEOUT_S = 180.0
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "L4"}, num_cards=1)
|
||||
@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,
|
||||
@@ -99,7 +99,7 @@ def test_voice_clone_streaming_001(omni_server, online_client) -> None:
|
||||
online_client.send_audio_speech_request(request_config, request_num=get_max_batch_size("few"))
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "L4"}, num_cards=1)
|
||||
@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,
|
||||
@@ -127,7 +127,7 @@ def test_response_format_001(omni_server, online_client) -> None:
|
||||
online_client.send_audio_speech_request(request_config)
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "L4"}, num_cards=1)
|
||||
@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``.
|
||||
|
||||
@@ -60,7 +60,7 @@ tts_server_params = [
|
||||
]
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "L4"}, num_cards=1)
|
||||
@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=1)
|
||||
@pytest.mark.parametrize("omni_server", tts_server_params, indirect=True)
|
||||
def test_voice_001(omni_server, online_client) -> None:
|
||||
"""
|
||||
@@ -92,7 +92,7 @@ def test_voice_001(omni_server, online_client) -> None:
|
||||
raise
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "L4"}, num_cards=1)
|
||||
@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=1)
|
||||
@pytest.mark.parametrize("omni_server", tts_server_params, indirect=True)
|
||||
def test_voice_002(omni_server, online_client) -> None:
|
||||
"""
|
||||
@@ -115,7 +115,7 @@ def test_voice_002(omni_server, online_client) -> None:
|
||||
online_client.send_audio_speech_request(request_config)
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "L4"}, num_cards=1)
|
||||
@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=1)
|
||||
@pytest.mark.parametrize("omni_server", tts_server_params, indirect=True)
|
||||
def test_voice_003(omni_server, online_client) -> None:
|
||||
"""
|
||||
@@ -138,7 +138,7 @@ def test_voice_003(omni_server, online_client) -> None:
|
||||
online_client.send_audio_speech_request(request_config)
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "L4"}, num_cards=1)
|
||||
@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=1)
|
||||
@pytest.mark.parametrize("omni_server", tts_server_params, indirect=True)
|
||||
def test_sample_rate_001(omni_server, online_client) -> None:
|
||||
"""The speech API resamples Qwen3-TTS output from native 24 kHz to 8 kHz."""
|
||||
@@ -155,7 +155,7 @@ def test_sample_rate_001(omni_server, online_client) -> None:
|
||||
online_client.send_audio_speech_request(request_config)
|
||||
|
||||
|
||||
@hardware_test(res={"cuda": "L4"}, num_cards=1)
|
||||
@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=1)
|
||||
@pytest.mark.parametrize("omni_server", tts_server_params, indirect=True)
|
||||
def test_language_001(omni_server, online_client) -> None:
|
||||
"""
|
||||
|
||||
@@ -21,8 +21,8 @@ pytestmark = [pytest.mark.diffusion, pytest.mark.full_model]
|
||||
|
||||
T2I_PROMPT = "A photo of a cat sitting on a laptop keyboard, digital art style."
|
||||
NEGATIVE_PROMPT = "blurry, low quality"
|
||||
SINGLE_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": "H100"})
|
||||
PARALLEL_FEATURE_MARKS = hardware_marks(res={"cuda": "H100"}, num_cards=2)
|
||||
SINGLE_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": ["H100", "B200"]})
|
||||
PARALLEL_FEATURE_MARKS = hardware_marks(res={"cuda": ["H100", "B200"]}, num_cards=2)
|
||||
|
||||
MODEL_IMAGE = "Qwen/Qwen-Image"
|
||||
MODEL_2512 = "Qwen/Qwen-Image-2512"
|
||||
|
||||
@@ -1,5 +1,5 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
|
||||
|
||||
"""
|
||||
Online serving smoke for SANA-WM image-to-video (first-frame I2V via ``/v1/videos``).
|
||||
|
||||
@@ -30,8 +30,8 @@ PROMPT = "Two anthropomorphic cats in comfy boxing gear and bright gloves fight
|
||||
NEGATIVE_PROMPT = "low quality, blurry, distorted face, extra limbs, bad anatomy, watermark, logo, text, ugly, deformed, mutated, jpeg artifacts"
|
||||
|
||||
# CUDA marks (original matrix, unchanged)
|
||||
CUDA_SINGLE_CARD_MARKS = hardware_marks(res={"cuda": "H100"})
|
||||
CUDA_PARALLEL_MARKS = hardware_marks(res={"cuda": "H100"}, num_cards=2)
|
||||
CUDA_SINGLE_CARD_MARKS = hardware_marks(res={"cuda": ["H100", "B200"]})
|
||||
CUDA_PARALLEL_MARKS = hardware_marks(res={"cuda": ["H100", "B200"]}, num_cards=2)
|
||||
|
||||
# NPU marks
|
||||
NPU_TWO_CARD_MARKS = hardware_marks(res={"npu": "A2"}, num_cards=2)
|
||||
|
||||
@@ -19,7 +19,7 @@ import pytest
|
||||
from tests.helpers.mark import hardware_marks
|
||||
from tests.helpers.runtime import OmniServer, OmniServerParams, OnlineOmniClient
|
||||
|
||||
pytestmark = [pytest.mark.diffusion, pytest.mark.full_model]
|
||||
pytestmark = [pytest.mark.diffusion, pytest.mark.slow]
|
||||
|
||||
MODEL = "Tongyi-MAI/Z-Image-Turbo"
|
||||
PROMPT = "A high-detail studio photo of an orange tabby cat sitting on a laptop keyboard."
|
||||
|
||||
Reference in new issue
Block a user