[CI] drop ascend scheduler test (#4582)

let' drop ascend scheduler test first to ensure all function works
without it.


- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-12-01 20:33:50 +08:00
committed by GitHub
parent 203b4e6777
commit 27b09ca9b9
28 changed files with 53 additions and 376 deletions

View File

@@ -20,7 +20,6 @@
Run `pytest tests/test_offline_inference.py`.
"""
import pytest
from vllm import SamplingParams
from vllm.assets.audio import AudioAsset
from vllm.assets.image import ImageAsset
@@ -55,40 +54,6 @@ def test_multimodal_vl(prompt_template):
assert output_str, "Generated output should not be empty."
@pytest.mark.skip(reason="This e2e test will stuck in multi-batch scenario. "
"Add this back after fixing the issue.")
def test_multimodal_ascend_scheduler(prompt_template):
image = ImageAsset("cherry_blossom") \
.pil_image.convert("RGB")
img_questions = [
"What is the content of this image?",
"Describe the content of this image in detail.",
"What's in the image?",
"Where is this image taken?",
]
images = [image] * len(img_questions)
prompts = prompt_template(img_questions)
with VllmRunner("Qwen/Qwen2.5-VL-3B-Instruct",
max_model_len=4096,
additional_config={
'ascend_scheduler_config': {
'enabled': True,
},
},
mm_processor_kwargs={
"min_pixels": 28 * 28,
"max_pixels": 1280 * 28 * 28,
"fps": 1,
},
enforce_eager=True) as vllm_model:
outputs = vllm_model.generate_greedy(prompts=prompts,
images=images,
max_tokens=64)
assert len(outputs) == len(prompts)
for _, output_str in outputs:
assert output_str, "Generated output should not be empty."
def test_multimodal_audio():
audio_prompt = "".join([
f"Audio {idx+1}: <|audio_bos|><|AUDIO|><|audio_eos|>\n"