A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[CI/Build] Split out mm processor tests (#23260)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Aug 21, 2025 at 11:05 UTC
2461d9e562e5852555c76e0dbed06979f9c6c688
3 files changed
+16
-8
.buildkite/test-pipeline.yaml
+11
-4
@@ -545,6 +545,15 @@ steps:
545
commands:
546
- pytest -v -s models/language/pooling -m 'not core_model'
547
548
+- label: Multi-Modal Processor Test
549
+ source_file_dependencies:
550
+ - vllm/
551
+ - tests/models/multimodal
552
+ commands:
553
+ - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
554
+ - pytest -v -s models/multimodal/processing --ignore models/multimodal/processing/test_tensor_schema.py
555
+ - pytest -v -s models/multimodal/processing/test_tensor_schema.py
556
+
557
- label: Multi-Modal Models Test (Standard)
558
mirror_hardwares: [amdexperimental]
559
torch_nightly: true
@@ -554,9 +563,7 @@ steps:
563
commands:
564
- pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
565
- pip freeze | grep -E 'torch'
557
- - pytest -v -s models/multimodal/processing
558
- - pytest -v -s --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/test_tensor_schema.py models/multimodal -m core_model
559
- - pytest -v -s models/multimodal/test_tensor_schema.py -m core_model # Needs mp_method="spawn"
566
+ - pytest -v -s models/multimodal -m core_model --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/processing
567
- cd .. && pytest -v -s tests/models/multimodal/generation/test_whisper.py -m core_model # Otherwise, mp_method="spawn" doesn't work
568
569
- label: Multi-Modal Models Test (Extended) 1
@@ -567,7 +574,7 @@ steps:
574
- tests/models/multimodal
575
commands:
576
- pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
570
- - pytest -v -s --ignore models/multimodal/generation/test_common.py --ignore models/multimodal/processing models/multimodal -m 'not core_model'
577
+ - pytest -v -s models/multimodal -m 'not core_model' --ignore models/multimodal/generation/test_common.py --ignore models/multimodal/processing
578
579
- label: Multi-Modal Models Test (Extended) 2
580
mirror_hardwares: [amdexperimental]
tests/models/multimodal/processing/test_tensor_schema.py
renamed
+3
-4
@@ -24,9 +24,9 @@ from vllm.utils import GiB_bytes, is_list_of, set_default_torch_num_threads
24
from vllm.v1.core.kv_cache_utils import get_kv_cache_config
25
from vllm.v1.engine.core import EngineCore as V1EngineCore
26
27
-from ...conftest import VllmRunner
28
-from ..registry import _MULTIMODAL_EXAMPLE_MODELS, HF_EXAMPLE_MODELS
29
-from ..utils import dummy_hf_overrides
27
+from ....conftest import VllmRunner
28
+from ...registry import _MULTIMODAL_EXAMPLE_MODELS, HF_EXAMPLE_MODELS
29
+from ...utils import dummy_hf_overrides
30
31
ARCH_TO_SKIP = {
32
"MolmoForCausalLM": "incompatible requirements",
@@ -147,7 +147,6 @@ def get_model_id_to_test(
147
return filtered_results
148
149
150
-@pytest.mark.core_model
150
@pytest.mark.parametrize(
151
"model_arch, model_id",
152
get_model_id_to_test(_MULTIMODAL_EXAMPLE_MODELS.keys()))
vllm/model_executor/models/cohere2_vision.py
+2
@@ -170,6 +170,8 @@ class Cohere2VisionProcessingInfo(BaseProcessingInfo):
170
171
# The current implementation of get_number_of_image_patches
172
# is incorrect, so we patch it here.
173
+ # TODO: Revert once
174
+ # https://github.com/huggingface/transformers/pull/40312 is released.
175
# return image_processor.get_number_of_image_patches(image_height,
176
# image_width, {})
177