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

164
tests/models/multimodal/generation/vlm_utils/builders.py
View File

@@ -1,7 +1,7 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Helpers for building inputs that can be leveraged for different test types.
"""
"""Helpers for building inputs that can be leveraged for different test types."""
from collections.abc import Iterable
from pathlib import PosixPath
from typing import Callable, Optional, Union
@@ -10,20 +10,30 @@ import torch
from vllm.multimodal.audio import AudioResampler
from vllm.multimodal.image import rescale_image_size
from vllm.multimodal.video import (rescale_video_size, resize_video,
sample_frames_from_video)
from vllm.multimodal.video import (
rescale_video_size,
resize_video,
sample_frames_from_video,
)
from .....conftest import AudioTestAssets, ImageTestAssets, VideoTestAssets
from .types import (SINGLE_AUDIO_BASE_PROMPT, SINGLE_IMAGE_BASE_PROMPTS,
TEST_AUDIO_PLACEHOLDER, TEST_IMG_PLACEHOLDER,
TEST_VIDEO_PLACEHOLDER, VIDEO_BASE_PROMPT,
ImageSizeWrapper, PromptWithMultiModalInput, SizeType,
VLMTestInfo)
from .types import (
SINGLE_AUDIO_BASE_PROMPT,
SINGLE_IMAGE_BASE_PROMPTS,
TEST_AUDIO_PLACEHOLDER,
TEST_IMG_PLACEHOLDER,
TEST_VIDEO_PLACEHOLDER,
VIDEO_BASE_PROMPT,
ImageSizeWrapper,
PromptWithMultiModalInput,
SizeType,
VLMTestInfo,
)
def replace_test_placeholder(prompt: str, mm_idx_to_prompt: Callable[[int],
str],
test_placeholder: str) -> str:
def replace_test_placeholder(
prompt: str, mm_idx_to_prompt: Callable[[int], str], test_placeholder: str
) -> str:
"""Given a prompt, replaces each test placeholder with the
model-specific tag.
"""
@@ -35,11 +45,13 @@ def replace_test_placeholder(prompt: str, mm_idx_to_prompt: Callable[[int],
return img_prompt
def get_model_prompts(base_prompts: Iterable[str],
img_idx_to_prompt: Optional[Callable[[int], str]],
video_idx_to_prompt: Optional[Callable[[int], str]],
audio_idx_to_prompt: Optional[Callable[[int], str]],
prompt_formatter: Callable[[str], str]) -> list[str]:
def get_model_prompts(
base_prompts: Iterable[str],
img_idx_to_prompt: Optional[Callable[[int], str]],
video_idx_to_prompt: Optional[Callable[[int], str]],
audio_idx_to_prompt: Optional[Callable[[int], str]],
prompt_formatter: Callable[[str], str],
) -> list[str]:
"""Given a model-agnostic base prompt and test configuration for a model(s)
to be tested, update the media placeholders and apply the prompt formatting
to get the test prompt string for this model.
@@ -56,19 +68,19 @@ def get_model_prompts(base_prompts: Iterable[str],
# Replace the multimodal placeholders in the base prompt with
# the correct ones for the model that we are testing
if img_idx_to_prompt:
base_prompt = replace_test_placeholder(base_prompt,
img_idx_to_prompt,
TEST_IMG_PLACEHOLDER)
base_prompt = replace_test_placeholder(
base_prompt, img_idx_to_prompt, TEST_IMG_PLACEHOLDER
)
if video_idx_to_prompt:
base_prompt = replace_test_placeholder(base_prompt,
video_idx_to_prompt,
TEST_VIDEO_PLACEHOLDER)
base_prompt = replace_test_placeholder(
base_prompt, video_idx_to_prompt, TEST_VIDEO_PLACEHOLDER
)
if audio_idx_to_prompt:
base_prompt = replace_test_placeholder(base_prompt,
audio_idx_to_prompt,
TEST_AUDIO_PLACEHOLDER)
base_prompt = replace_test_placeholder(
base_prompt, audio_idx_to_prompt, TEST_AUDIO_PLACEHOLDER
)
# Apply the prompt formatter to wrap the base prompt with
# the correct media placeholders to get the model test prompt
@@ -84,14 +96,15 @@ def build_single_image_inputs_from_test_info(
tmp_path: Optional[PosixPath] = None,
) -> list[PromptWithMultiModalInput]:
if test_info.prompt_formatter is None:
raise ValueError(
"Prompt formatter must be set to build single image inputs")
raise ValueError("Prompt formatter must be set to build single image inputs")
model_prompts = get_model_prompts(test_info.single_image_prompts,
test_info.img_idx_to_prompt,
test_info.video_idx_to_prompt,
test_info.audio_idx_to_prompt,
test_info.prompt_formatter)
model_prompts = get_model_prompts(
test_info.single_image_prompts,
test_info.img_idx_to_prompt,
test_info.video_idx_to_prompt,
test_info.audio_idx_to_prompt,
test_info.prompt_formatter,
)
# For models that require a local path / URL encoded in the image; export
# assets and encode into tmp_path for this test. This should be avoided
@@ -110,8 +123,8 @@ def build_single_image_inputs_from_test_info(
def build_single_image_inputs(
images, model_prompts,
size_wrapper: ImageSizeWrapper) -> list[PromptWithMultiModalInput]:
images, model_prompts, size_wrapper: ImageSizeWrapper
) -> list[PromptWithMultiModalInput]:
# For every image / prompt pair, get a pair containing two lists of
# length size_factors, where the first contains duplicates of the model
# prompt [str], and the second contains copies of the image after being
@@ -125,7 +138,8 @@ def build_single_image_inputs(
apply_image_size_scaling(image, size, size_wrapper.type)
for size in size_wrapper.data
],
) for image, prompt in zip(images, model_prompts)
)
for image, prompt in zip(images, model_prompts)
]
@@ -136,14 +150,15 @@ def build_multi_image_inputs_from_test_info(
tmp_path: Optional[PosixPath] = None,
) -> list[PromptWithMultiModalInput]:
if test_info.prompt_formatter is None:
raise ValueError(
"Prompt formatter must be set to build multi image inputs")
raise ValueError("Prompt formatter must be set to build multi image inputs")
model_prompts = get_model_prompts([test_info.multi_image_prompt],
test_info.img_idx_to_prompt,
test_info.video_idx_to_prompt,
test_info.audio_idx_to_prompt,
test_info.prompt_formatter)
model_prompts = get_model_prompts(
[test_info.multi_image_prompt],
test_info.img_idx_to_prompt,
test_info.video_idx_to_prompt,
test_info.audio_idx_to_prompt,
test_info.prompt_formatter,
)
if test_info.prompt_path_encoder is not None:
if tmp_path is None:
@@ -164,16 +179,20 @@ def build_multi_image_inputs_from_test_info(
def build_multi_image_inputs(
image_lists, model_prompts,
size_wrapper: ImageSizeWrapper) -> list[PromptWithMultiModalInput]:
image_lists, model_prompts, size_wrapper: ImageSizeWrapper
) -> list[PromptWithMultiModalInput]:
return [
PromptWithMultiModalInput(
prompts=[prompt for _ in size_wrapper.data],
image_data=[[
apply_image_size_scaling(image, size, size_wrapper.type)
for image in images
] for size in size_wrapper.data],
) for images, prompt in zip(image_lists, model_prompts)
image_data=[
[
apply_image_size_scaling(image, size, size_wrapper.type)
for image in images
]
for size in size_wrapper.data
],
)
for images, prompt in zip(image_lists, model_prompts)
]
@@ -185,10 +204,10 @@ def build_embedding_inputs_from_test_info(
# These conditions will always be true if invoked through filtering,
# but we still check them in case this is ever called directly
if test_info.prompt_formatter is None:
raise ValueError(
"Prompt formatter must be set to build image embedding inputs")
if size_wrapper.type != SizeType.SIZE_FACTOR or not \
all(factor == 1.0 for factor in size_wrapper.data):
raise ValueError("Prompt formatter must be set to build image embedding inputs")
if size_wrapper.type != SizeType.SIZE_FACTOR or not all(
factor == 1.0 for factor in size_wrapper.data
):
raise ValueError("Embedding tests require constant (1.0) size factors")
if test_info.convert_assets_to_embeddings is None:
raise ValueError("No conversion func for getting embeddings found")
@@ -209,8 +228,7 @@ def build_embedding_inputs_from_test_info(
assert len(images) == len(model_prompts)
inputs = build_single_image_inputs(images, model_prompts, size_wrapper)
vllm_embeddings = build_single_image_inputs(embeds, model_prompts,
size_wrapper)
vllm_embeddings = build_single_image_inputs(embeds, model_prompts, size_wrapper)
return inputs, vllm_embeddings
@@ -235,21 +253,22 @@ def build_video_inputs_from_test_info(
for asset in video_assets
]
video_scaler = (resize_video if size_wrapper.type == SizeType.FIXED_SIZE
else rescale_video_size)
video_scaler = (
resize_video if size_wrapper.type == SizeType.FIXED_SIZE else rescale_video_size
)
return [
PromptWithMultiModalInput(
prompts=[prompt for _ in size_wrapper.data],
video_data=[
video_scaler(video, size) for size in size_wrapper.data
],
) for video, prompt in zip(sampled_vids, model_prompts)
video_data=[video_scaler(video, size) for size in size_wrapper.data],
)
for video, prompt in zip(sampled_vids, model_prompts)
]
def apply_image_size_scaling(image, size: Union[float, tuple[int, int]],
size_type: SizeType):
def apply_image_size_scaling(
image, size: Union[float, tuple[int, int]], size_type: SizeType
):
"""Applies a size scaler to one image; this can be an image size factor,
which scales the image while maintaining the aspect ratio"""
# Special case for embeddings; if it's a tensor, it's only valid if we
@@ -285,13 +304,16 @@ def build_audio_inputs_from_test_info(
method="librosa",
)
audios = [asset.audio_and_sample_rate for asset in audio_assets]
resampled_audios = [(
resampler.resample(
audio,
orig_sr=sr,
),
int(resampler.target_sr),
) for audio, sr in audios]
resampled_audios = [
(
resampler.resample(
audio,
orig_sr=sr,
),
int(resampler.target_sr),
)
for audio, sr in audios
]
return [
PromptWithMultiModalInput(