[CI][CPU] Add Qwen2-VL multimodal tests for CPU backend and fix incompatibilities (#48072)
Signed-off-by: Zhenhui Zhao <zhenhui.zhao@intel.com>
This commit is contained in:
@@ -141,9 +141,22 @@ steps:
|
|||||||
commands:
|
commands:
|
||||||
- |
|
- |
|
||||||
bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 45m "
|
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
|
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"
|
- label: "Arm CPU Test"
|
||||||
depends_on: []
|
depends_on: []
|
||||||
soft_fail: false
|
soft_fail: false
|
||||||
|
|||||||
@@ -5,6 +5,7 @@ import pytest
|
|||||||
|
|
||||||
from vllm.assets.image import ImageAsset
|
from vllm.assets.image import ImageAsset
|
||||||
from vllm.multimodal.video import sample_frames_from_video
|
from vllm.multimodal.video import sample_frames_from_video
|
||||||
|
from vllm.platforms import current_platform
|
||||||
|
|
||||||
from ....conftest import VIDEO_ASSETS
|
from ....conftest import VIDEO_ASSETS
|
||||||
|
|
||||||
@@ -52,11 +53,15 @@ def _encoder_cudagraph_config(*, max_vision_items: int) -> dict:
|
|||||||
|
|
||||||
@pytest.mark.core_model
|
@pytest.mark.core_model
|
||||||
@pytest.mark.parametrize("model", models)
|
@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("num_frames", [16])
|
||||||
@pytest.mark.parametrize("dtype", [target_dtype])
|
@pytest.mark.parametrize("dtype", [target_dtype])
|
||||||
@pytest.mark.parametrize("max_tokens", [128])
|
@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(
|
def test_qwen2_5_vl_evs_functionality(
|
||||||
vllm_runner,
|
vllm_runner,
|
||||||
video_assets,
|
video_assets,
|
||||||
@@ -109,11 +114,15 @@ def test_qwen2_5_vl_evs_functionality(
|
|||||||
|
|
||||||
@pytest.mark.core_model
|
@pytest.mark.core_model
|
||||||
@pytest.mark.parametrize("model", models)
|
@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("num_frames", [16])
|
||||||
@pytest.mark.parametrize("dtype", [target_dtype])
|
@pytest.mark.parametrize("dtype", [target_dtype])
|
||||||
@pytest.mark.parametrize("max_tokens", [128])
|
@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(
|
def test_qwen2_5_vl_evs_batched_videos(
|
||||||
vllm_runner,
|
vllm_runner,
|
||||||
video_assets,
|
video_assets,
|
||||||
@@ -174,7 +183,9 @@ def test_qwen2_5_vl_evs_batched_videos(
|
|||||||
@pytest.mark.parametrize("model", models)
|
@pytest.mark.parametrize("model", models)
|
||||||
@pytest.mark.parametrize("dtype", [target_dtype])
|
@pytest.mark.parametrize("dtype", [target_dtype])
|
||||||
@pytest.mark.parametrize("max_tokens", [128])
|
@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(
|
def test_qwen2_5_vl_window_attention_image(
|
||||||
vllm_runner,
|
vllm_runner,
|
||||||
model,
|
model,
|
||||||
@@ -210,7 +221,9 @@ def test_qwen2_5_vl_window_attention_image(
|
|||||||
@pytest.mark.parametrize("model", models)
|
@pytest.mark.parametrize("model", models)
|
||||||
@pytest.mark.parametrize("dtype", [target_dtype])
|
@pytest.mark.parametrize("dtype", [target_dtype])
|
||||||
@pytest.mark.parametrize("max_tokens", [128])
|
@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(
|
def test_qwen2_5_vl_window_attention_image_batch(
|
||||||
vllm_runner,
|
vllm_runner,
|
||||||
model,
|
model,
|
||||||
|
|||||||
Reference in New Issue
Block a user