iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[CI][CPU] Add Qwen2-VL multimodal tests for CPU backend and fix incompatibilities (#48072)

Browse Source 
Signed-off-by: Zhenhui Zhao <zhenhui.zhao@intel.com>
This commit is contained in:
zhao, zhenhui
2026-07-12 12:30:34 +08:00
committed by GitHub
parent 9a48eef89a
commit 8e981630c9
2 changed files with 33 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.buildkite/hardware_tests/cpu.yaml
+14 -1
View File
@@ -141,9 +141,22 @@ steps:
commands:
- |
bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 45m "
pytest -x -v -s tests/models/multimodal/generation --ignore=tests/models/multimodal/generation/test_pixtral.py -m cpu_model --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB"
pytest -x -v -s tests/models/multimodal/generation --ignore=tests/models/multimodal/generation/test_pixtral.py --ignore=tests/models/multimodal/generation/test_qwen2_5_vl.py -m cpu_model --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB"
parallelism: 4
- label: CPU-Qwen2.5-VL Multimodal Tests
depends_on: []
device: intel_cpu
no_plugin: true
source_file_dependencies:
# - vllm/
- vllm/model_executor/layers/rotary_embedding
- tests/models/multimodal/generation/
commands:
- |
bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 40m "
VLLM_CI_ENV=0 pytest -x -v -s tests/models/multimodal/generation/test_qwen2_5_vl.py"
- label: "Arm CPU Test"
depends_on: []
soft_fail: false
tests/models/multimodal/generation/test_qwen2_5_vl.py
+19 -6
View File
@@ -5,6 +5,7 @@ import pytest
from vllm.assets.image import ImageAsset
from vllm.multimodal.video import sample_frames_from_video
from vllm.platforms import current_platform
from ....conftest import VIDEO_ASSETS
@@ -52,11 +53,15 @@ def _encoder_cudagraph_config(*, max_vision_items: int) -> dict:
@pytest.mark.core_model
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize("video_pruning_rate", [0.0, 0.75])
@pytest.mark.parametrize(
"video_pruning_rate", [0.0] if current_platform.is_cpu() else [0.0, 0.75]
)
@pytest.mark.parametrize("num_frames", [16])
@pytest.mark.parametrize("dtype", [target_dtype])
@pytest.mark.parametrize("max_tokens", [128])
@pytest.mark.parametrize("use_bytecode_hook", [True, False])
@pytest.mark.parametrize(
"use_bytecode_hook", [True] if current_platform.is_cpu() else [True, False]
)
def test_qwen2_5_vl_evs_functionality(
vllm_runner,
video_assets,
@@ -109,11 +114,15 @@ def test_qwen2_5_vl_evs_functionality(
@pytest.mark.core_model
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize("video_pruning_rate", [0.0, 0.75])
@pytest.mark.parametrize(
"video_pruning_rate", [0.0] if current_platform.is_cpu() else [0.0, 0.75]
)
@pytest.mark.parametrize("num_frames", [16])
@pytest.mark.parametrize("dtype", [target_dtype])
@pytest.mark.parametrize("max_tokens", [128])
@pytest.mark.parametrize("use_bytecode_hook", [True, False])
@pytest.mark.parametrize(
"use_bytecode_hook", [True] if current_platform.is_cpu() else [True, False]
)
def test_qwen2_5_vl_evs_batched_videos(
vllm_runner,
video_assets,
@@ -174,7 +183,9 @@ def test_qwen2_5_vl_evs_batched_videos(
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize("dtype", [target_dtype])
@pytest.mark.parametrize("max_tokens", [128])
@pytest.mark.parametrize("use_bytecode_hook", [True, False])
@pytest.mark.parametrize(
"use_bytecode_hook", [True] if current_platform.is_cpu() else [True, False]
)
def test_qwen2_5_vl_window_attention_image(
vllm_runner,
model,
@@ -210,7 +221,9 @@ def test_qwen2_5_vl_window_attention_image(
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize("dtype", [target_dtype])
@pytest.mark.parametrize("max_tokens", [128])
@pytest.mark.parametrize("use_bytecode_hook", [True, False])
@pytest.mark.parametrize(
"use_bytecode_hook", [True] if current_platform.is_cpu() else [True, False]
)
def test_qwen2_5_vl_window_attention_image_batch(
vllm_runner,
model,