[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:
authored and GitHub committed 2026-09-14 14:27:10 +08:00
1 parent 58adeec05f
commit 64903b8e30
78 files changed
+437 -321

No files matched your search

@@ -83,7 +83,7 @@ def _as_uint8_frames(video: np.ndarray) -> np.ndarray:
_MAX_DIFFERING_FRACTION = 0.35
@hardware_test(res={"cuda": "H100"}, num_cards=1)
@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards=1)
def test_device_postprocess_produces_equivalent_frames() -> None:
float_path = _generate(False)
device_path = _generate(True)
@@ -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 tests for dots.tts native AR offline inference."""
from collections.abc import Mapping
@@ -246,7 +246,7 @@ def _transcribe_for(wav: torch.Tensor, tmp_dir: str, name: str, accepts) -> str:
pytest.param(
"The quick brown fox jumps over the lazy dog.",
None,
marks=pytest.mark.full_model,
marks=pytest.mark.slow,
id="second_text",
),
],
@@ -274,7 +274,7 @@ def test_gepard_offline_zero_shot(omni_runner, run_level: str, text: str, keywor
assert keyword in transcript, f"expected {keyword!r} in transcript, got {transcript!r}"
@pytest.mark.full_model
@pytest.mark.slow
@hardware_test(res={"cuda": "L4"}, num_cards=1)
def test_gepard_offline_long_text_skips_repetition(omni_runner, run_level: str) -> None:
"""The other branch of the prompt layout, end to end.
@@ -57,7 +57,7 @@ def omni() -> Generator[Omni, None, None]:
yield runner.omni
@hardware_test(res={"cuda": "H100"}, num_cards=4)
@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards=4)
def test_i2t_generates_text(omni: Omni) -> None:
"""Verify I2T output's first 20 token IDs match the HF greedy baseline."""
# Solid-color image keeps the input self-contained and reproducible.
@@ -47,7 +47,7 @@ pytestmark = [
]
@hardware_test(res={"cuda": "H100"}, num_cards=1)
@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards=1)
def test_lingbot_world_v2_tp1_one_block(omni_runner_handler: OmniRunnerHandler) -> None:
"""Load the real checkpoint and generate one causal block on one GPU."""
@@ -1,7 +1,8 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
from collections.abc import Generator
from typing import Any
import numpy as np
import pytest
@@ -105,15 +106,17 @@ def _generate_avatar_frames(
num_frames: int = 5,
sampling_extra_args: dict | None = None,
):
extra_info: dict[str, Any] = dict(additional_information or {})
extra_info.update({"stage": stage, "resolution": "480p"})
multi_modal_data: dict[str, Any] = {"audio": audio_path}
if image_path is not None:
multi_modal_data["image"] = image_path
prompt = {
"prompt": PROMPT,
"negative_prompt": NEGATIVE_PROMPT,
"multi_modal_data": {"audio": audio_path},
"additional_information": additional_information or {},
"multi_modal_data": multi_modal_data,
"additional_information": extra_info,
}
prompt["additional_information"].update({"stage": stage, "resolution": "480p"})
if image_path is not None:
prompt["multi_modal_data"]["image"] = image_path
outputs = omni_runner.generate(
[prompt],
@@ -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 test for Qwen-Image AutoRound W4A16 quantized inference.
Verifies that the W4A16 quantized Qwen-Image checkpoint loads end-to-end
@@ -47,8 +47,9 @@ def _first_request_images(outputs) -> list:
return req_out.images
@pytest.mark.full_model
@pytest.mark.diffusion
@hardware_test(res={"cuda": "L4"})
@hardware_test(res={"cuda": ["L4", "B200"]})
def test_qwen_image_autoround_w4a16_load():
"""Load the W4A16 quantized Qwen-Image model and run a minimal generation.
@@ -68,8 +69,9 @@ def test_qwen_image_autoround_w4a16_load():
assert arr.std() > 1.0, "Generated image appears blank (std ≈ 0)"
@pytest.mark.full_model
@pytest.mark.diffusion
@hardware_test(res={"cuda": "L4"})
@hardware_test(res={"cuda": ["L4", "B200"]})
def test_qwen_image_autoround_w4a16_generate():
"""Full generation: 512×512, 20 steps, CFG=5.0.
@@ -1,3 +1,6 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
"""E2E test for VoxCPM2 native AR offline inference."""
import os
@@ -156,7 +156,7 @@ def _generate_t2v_video(offline_client, prompt: str = "A cat sitting on a table"
# ------------------------------------------------------------------
@hardware_test(res={"cuda": "H100"})
@hardware_test(res={"cuda": ["H100", "B200"]})
@pytest.mark.parametrize("omni_runner", quant_i2v_params, indirect=True)
def test_wan22_i2v_autoround_w4a16_generates_video(omni_runner, offline_client):
"""Load the W4A16 quantized Wan2.2 I2V model and verify it produces a valid video."""
@@ -180,7 +180,7 @@ def test_wan22_i2v_autoround_w4a16_generates_video(omni_runner, offline_client):
# ------------------------------------------------------------------
@hardware_test(res={"cuda": "H100"})
@hardware_test(res={"cuda": ["H100", "B200"]})
@pytest.mark.parametrize("omni_runner", quant_t2v_params, indirect=True)
def test_wan22_t2v_autoround_w4a16_generates_video(omni_runner, offline_client):
"""Load the W4A16 quantized Wan2.2 T2V model and verify it produces a valid video."""
@@ -202,7 +202,7 @@ def test_wan22_t2v_autoround_w4a16_generates_video(omni_runner, offline_client):
# ------------------------------------------------------------------
@hardware_test(res={"cuda": "H100"})
@hardware_test(res={"cuda": ["H100", "B200"]})
@pytest.mark.parametrize("omni_runner", quant_i2v_params, indirect=True)
def test_wan22_i2v_autoround_w4a16_quant_peak(omni_runner, offline_client):
"""Measure peak GPU memory of W4A16 quantized I2V model."""
@@ -218,7 +218,7 @@ def test_wan22_i2v_autoround_w4a16_quant_peak(omni_runner, offline_client):
# ------------------------------------------------------------------
@hardware_test(res={"cuda": "H100"})
@hardware_test(res={"cuda": ["H100", "B200"]})
@pytest.mark.parametrize("omni_runner", baseline_i2v_params, indirect=True)
def test_wan22_i2v_autoround_w4a16_baseline_peak(omni_runner, offline_client):
"""Measure peak GPU memory of BF16 baseline I2V model."""
@@ -234,7 +234,7 @@ def test_wan22_i2v_autoround_w4a16_baseline_peak(omni_runner, offline_client):
# ------------------------------------------------------------------
@hardware_test(res={"cuda": "H100"})
@hardware_test(res={"cuda": ["H100", "B200"]})
def test_wan22_i2v_autoround_w4a16_memory_savings():
"""Assert quantized I2V model uses meaningfully less memory than BF16 baseline."""
quant_peak = _memory_results["quant_i2v"]
@@ -259,7 +259,7 @@ def test_wan22_i2v_autoround_w4a16_memory_savings():
# ------------------------------------------------------------------
@hardware_test(res={"cuda": "H100"})
@hardware_test(res={"cuda": ["H100", "B200"]})
@pytest.mark.parametrize("omni_runner", quant_t2v_params, indirect=True)
def test_wan22_t2v_autoround_w4a16_quant_peak(omni_runner, offline_client):
"""Measure peak GPU memory of W4A16 quantized T2V model."""
@@ -275,7 +275,7 @@ def test_wan22_t2v_autoround_w4a16_quant_peak(omni_runner, offline_client):
# ------------------------------------------------------------------
@hardware_test(res={"cuda": "H100"})
@hardware_test(res={"cuda": ["H100", "B200"]})
@pytest.mark.parametrize("omni_runner", baseline_t2v_params, indirect=True)
def test_wan22_t2v_autoround_w4a16_baseline_peak(omni_runner, offline_client):
"""Measure peak GPU memory of BF16 baseline T2V model."""
@@ -291,7 +291,7 @@ def test_wan22_t2v_autoround_w4a16_baseline_peak(omni_runner, offline_client):
# ------------------------------------------------------------------
@hardware_test(res={"cuda": "H100"})
@hardware_test(res={"cuda": ["H100", "B200"]})
def test_wan22_t2v_autoround_w4a16_memory_savings():
"""Assert quantized T2V model uses meaningfully less memory than BF16 baseline."""
quant_peak = _memory_results["quant_t2v"]