biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Convert formatting to use ruff instead of yapf + isort (#26247)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-05 15:06:22 +01:00
committed by GitHub
parent 17edd8a807
commit d6953beb91
1508 changed files with 115244 additions and 94146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

105
tests/models/multimodal/processing/test_tensor_schema.py
View File

@@ -9,23 +9,29 @@ from typing import Any, Union
import numpy as np
import pytest
import torch.nn as nn
from mistral_common.protocol.instruct.messages import (ImageChunk, TextChunk,
UserMessage)
from mistral_common.protocol.instruct.messages import ImageChunk, TextChunk, UserMessage
from mistral_common.protocol.instruct.request import ChatCompletionRequest
from PIL import Image
from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config
from vllm.config.multimodal import (AudioDummyOptions, BaseDummyOptions,
ImageDummyOptions, VideoDummyOptions)
from vllm.distributed import (cleanup_dist_env_and_memory,
init_distributed_environment,
initialize_model_parallel)
from vllm.config.multimodal import (
AudioDummyOptions,
BaseDummyOptions,
ImageDummyOptions,
VideoDummyOptions,
)
from vllm.distributed import (
cleanup_dist_env_and_memory,
init_distributed_environment,
initialize_model_parallel,
)
from vllm.model_executor.model_loader.utils import set_default_torch_dtype
from vllm.model_executor.models.interfaces import (SupportsMultiModal,
supports_multimodal)
from vllm.model_executor.models.interfaces import (
SupportsMultiModal,
supports_multimodal,
)
from vllm.multimodal import MULTIMODAL_REGISTRY, BatchedTensorInputs
from vllm.multimodal.processing import (BaseMultiModalProcessor,
InputProcessingContext)
from vllm.multimodal.processing import BaseMultiModalProcessor, InputProcessingContext
from vllm.multimodal.utils import group_mm_kwargs_by_modality
from vllm.transformers_utils.tokenizer import cached_tokenizer_from_config
from vllm.utils import is_list_of
@@ -48,13 +54,15 @@ REPO_ID_TO_SKIP = {
}
ImageInput = list[Image.Image]
VideoInput = Union[list[Image.Image], list[np.ndarray],
list[tuple[np.ndarray, dict[str, Any]]]]
VideoInput = Union[
list[Image.Image], list[np.ndarray], list[tuple[np.ndarray, dict[str, Any]]]
]
AudioInput = list[tuple[np.ndarray, int]]
def _resize_data(_data: Union[Image.Image, np.ndarray],
size_factor: float) -> Union[Image.Image, np.ndarray]:
def _resize_data(
_data: Union[Image.Image, np.ndarray], size_factor: float
) -> Union[Image.Image, np.ndarray]:
assert size_factor <= 1, "Size factor must be less than 1"
# Image input
if isinstance(_data, Image.Image):
@@ -74,20 +82,18 @@ def _resize_data(_data: Union[Image.Image, np.ndarray],
return _data[..., :T, :H, :W, :C]
# Audio input
elif isinstance(_data, np.ndarray) and _data.ndim == 1:
return _data[:int(len(_data) * size_factor)]
return _data[: int(len(_data) * size_factor)]
raise AssertionError("This line should be unreachable.")
def resize_mm_data(
data: Union[ImageInput, VideoInput, AudioInput],
size_factors: tuple[float,
...]) -> Union[ImageInput, VideoInput, AudioInput]:
size_factors = size_factors[:len(data)]
data: Union[ImageInput, VideoInput, AudioInput], size_factors: tuple[float, ...]
) -> Union[ImageInput, VideoInput, AudioInput]:
size_factors = size_factors[: len(data)]
if is_list_of(data, (Image.Image, np.ndarray, list)):
return [_resize_data(d, s) for d, s in zip(data, size_factors)]
elif is_list_of(data, tuple):
return [(_resize_data(d, s), meta)
for (d, meta), s in zip(data, size_factors)]
return [(_resize_data(d, s), meta) for (d, meta), s in zip(data, size_factors)]
raise ValueError("Unsupported multimodal data type.")
@@ -116,12 +122,16 @@ def create_batched_mm_kwargs(
# Mistral chat outputs tokens directly, rather than text prompts
if model_config.tokenizer_mode == "mistral":
images = resized_mm_data.get("image", [])
request = ChatCompletionRequest(messages=[
UserMessage(content=[
TextChunk(text=""),
*(ImageChunk(image=image) for image in images),
]),
])
request = ChatCompletionRequest(
messages=[
UserMessage(
content=[
TextChunk(text=""),
*(ImageChunk(image=image) for image in images),
]
),
]
)
tokenizer = processing_info.get_tokenizer()
res = tokenizer.mistral.encode_chat_completion(request)
prompt = res.tokens
@@ -133,10 +143,7 @@ def create_batched_mm_kwargs(
hf_processor_mm_kwargs=processor_inputs.hf_processor_mm_kwargs,
tokenization_kwargs=processor_inputs.tokenization_kwargs,
)["mm_kwargs"].require_data()
items = [
item for modality in supported_mm_limits
for item in mm_kwargs[modality]
]
items = [item for modality in supported_mm_limits for item in mm_kwargs[modality]]
return group_mm_kwargs_by_modality(
items,
merge_by_field_config=model_cls.merge_by_field_config,
@@ -167,15 +174,17 @@ def initialize_dummy_model(
cleanup_dist_env_and_memory()
def get_model_id_to_test(
model_arch_list: Iterable[str]) -> list[tuple[str, str]]:
def get_model_id_to_test(model_arch_list: Iterable[str]) -> list[tuple[str, str]]:
filtered_results = []
for model_arch in model_arch_list:
model_info = HF_EXAMPLE_MODELS.get_hf_info(model_arch)
if model_info.extras and model_arch in ARCH_NEEDS_EXTRAS:
available_repos = list(
map(lambda model_id: (model_arch, model_id),
[model_info.default, *model_info.extras.values()]))
map(
lambda model_id: (model_arch, model_id),
[model_info.default, *model_info.extras.values()],
)
)
filtered_results.extend(available_repos)
else:
filtered_results.append((model_arch, model_info.default))
@@ -183,8 +192,8 @@ def get_model_id_to_test(
@pytest.mark.parametrize(
"model_arch, model_id",
get_model_id_to_test(_MULTIMODAL_EXAMPLE_MODELS.keys()))
"model_arch, model_id", get_model_id_to_test(_MULTIMODAL_EXAMPLE_MODELS.keys())
)
def test_model_tensor_schema(model_arch: str, model_id: str):
if model_arch in ARCH_TO_SKIP:
pytest.skip(f"Skipping {model_arch} due to {ARCH_TO_SKIP[model_arch]}")
@@ -193,12 +202,13 @@ def test_model_tensor_schema(model_arch: str, model_id: str):
model_info = HF_EXAMPLE_MODELS.get_hf_info(model_arch)
model_info.check_available_online(on_fail="skip")
model_info.check_transformers_version(on_fail="skip",
check_max_version=False)
model_info.check_transformers_version(on_fail="skip", check_max_version=False)
hf_overrides_fn = partial(dummy_hf_overrides,
model_arch=model_arch,
exist_overrides=model_info.hf_overrides)
hf_overrides_fn = partial(
dummy_hf_overrides,
model_arch=model_arch,
exist_overrides=model_info.hf_overrides,
)
model_config = ModelConfig(
model_id,
@@ -256,8 +266,11 @@ def test_model_tensor_schema(model_arch: str, model_id: str):
with initialize_dummy_model(model_cls, model_config) as model:
for modality, _, mm_kwargs in create_batched_mm_kwargs(
model_cls, model_config, processor):
model_cls, model_config, processor
):
for method_name in inputs_parse_methods:
print(f"Testing `{method_name}` with modality={modality} "
f"and mm_kwargs{list(mm_kwargs.keys())}")
print(
f"Testing `{method_name}` with modality={modality} "
f"and mm_kwargs{list(mm_kwargs.keys())}"
)
getattr(model, method_name)(modality=modality, **mm_kwargs)