e489ad7a21
- **Add SPDX license headers to python source files** - **Check for SPDX headers using pre-commit** commit 9d7ef44c3cfb72ca4c32e1c677d99259d10d4745 Author: Russell Bryant <rbryant@redhat.com> Date: Fri Jan 31 14:18:24 2025 -0500 Add SPDX license headers to python source files This commit adds SPDX license headers to python source files as recommended to the project by the Linux Foundation. These headers provide a concise way that is both human and machine readable for communicating license information for each source file. It helps avoid any ambiguity about the license of the code and can also be easily used by tools to help manage license compliance. The Linux Foundation runs license scans against the codebase to help ensure we are in compliance with the licenses of the code we use, including dependencies. Having these headers in place helps that tool do its job. More information can be found on the SPDX site: - https://spdx.dev/learn/handling-license-info/ Signed-off-by: Russell Bryant <rbryant@redhat.com> commit 5a1cf1cb3b80759131c73f6a9dddebccac039dea Author: Russell Bryant <rbryant@redhat.com> Date: Fri Jan 31 14:36:32 2025 -0500 Check for SPDX headers using pre-commit Signed-off-by: Russell Bryant <rbryant@redhat.com> --------- Signed-off-by: Russell Bryant <rbryant@redhat.com>
123 lines
4.2 KiB
Python
123 lines
4.2 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
|
|
from typing import List, Optional, Type
|
|
|
|
import pytest
|
|
import torch
|
|
|
|
from vllm.multimodal.image import rescale_image_size
|
|
|
|
from ....conftest import IMAGE_ASSETS, VllmRunner, _ImageAssets
|
|
from ...utils import check_logprobs_close
|
|
|
|
HF_IMAGE_PROMPTS = IMAGE_ASSETS.prompts({
|
|
"stop_sign":
|
|
"<|im_start|>User\n<image>\nWhat's the content in the center of the image?<|im_end|>\n<|im_start|>Assistant\n", # noqa: E501
|
|
"cherry_blossom":
|
|
"<|im_start|>User\n<image>\nWhat is the season?<|im_end|>\n<|im_start|>Assistant\n", # noqa: E501
|
|
})
|
|
|
|
|
|
def run_awq_test(
|
|
vllm_runner: Type[VllmRunner],
|
|
image_assets: _ImageAssets,
|
|
source_model: str,
|
|
quant_model: str,
|
|
*,
|
|
size_factors: List[float],
|
|
dtype: str,
|
|
max_tokens: int,
|
|
num_logprobs: int,
|
|
tensor_parallel_size: int,
|
|
distributed_executor_backend: Optional[str] = None,
|
|
):
|
|
images = [asset.pil_image for asset in image_assets]
|
|
|
|
inputs_per_image = [(
|
|
[prompt for _ in size_factors],
|
|
[rescale_image_size(image, factor) for factor in size_factors],
|
|
) for image, prompt in zip(images, HF_IMAGE_PROMPTS)]
|
|
|
|
# NOTE: take care of the order. run vLLM first, and then run HF.
|
|
# vLLM needs a fresh new process without cuda initialization.
|
|
# if we run HF first, the cuda initialization will be done and it
|
|
# will hurt multiprocessing backend with fork method (the default method).
|
|
|
|
# max_model_len should be greater than image_feature_size
|
|
with vllm_runner(source_model,
|
|
max_model_len=4096,
|
|
dtype=dtype,
|
|
tensor_parallel_size=tensor_parallel_size,
|
|
distributed_executor_backend=distributed_executor_backend,
|
|
enforce_eager=True) as vllm_model:
|
|
source_outputs_per_image = [
|
|
vllm_model.generate_greedy_logprobs(prompts,
|
|
max_tokens,
|
|
num_logprobs=num_logprobs,
|
|
images=images)
|
|
for prompts, images in inputs_per_image
|
|
]
|
|
|
|
with vllm_runner(quant_model,
|
|
quantization="awq",
|
|
max_model_len=4096,
|
|
dtype=dtype,
|
|
tensor_parallel_size=tensor_parallel_size,
|
|
distributed_executor_backend=distributed_executor_backend,
|
|
enforce_eager=True) as vllm_model:
|
|
quant_outputs_per_image = [
|
|
vllm_model.generate_greedy_logprobs(prompts,
|
|
max_tokens,
|
|
num_logprobs=num_logprobs,
|
|
images=images)
|
|
for prompts, images in inputs_per_image
|
|
]
|
|
|
|
for source_outputs, quant_outputs in zip(source_outputs_per_image,
|
|
quant_outputs_per_image):
|
|
# TODO: Check whether using original CLIPVisionModel can improve
|
|
# consistency against HF
|
|
check_logprobs_close(
|
|
outputs_0_lst=source_outputs,
|
|
outputs_1_lst=quant_outputs,
|
|
name_0="source",
|
|
name_1="awq",
|
|
)
|
|
|
|
|
|
@pytest.mark.quant_model
|
|
@pytest.mark.parametrize(
|
|
("source_model", "quant_model"),
|
|
[("OpenGVLab/InternVL2-2B", "OpenGVLab/InternVL2-2B-AWQ")],
|
|
)
|
|
@pytest.mark.parametrize(
|
|
"size_factors",
|
|
[
|
|
# No image
|
|
[],
|
|
# Single-scale
|
|
[1.0],
|
|
# Single-scale, batched
|
|
[1.0, 1.0, 1.0],
|
|
# Multi-scale
|
|
[0.25, 0.5, 1.0],
|
|
],
|
|
)
|
|
@pytest.mark.parametrize("dtype", ["half"])
|
|
@pytest.mark.parametrize("max_tokens", [128])
|
|
@pytest.mark.parametrize("num_logprobs", [5])
|
|
@torch.inference_mode()
|
|
def test_awq_models(vllm_runner, image_assets, source_model, quant_model,
|
|
size_factors, dtype, max_tokens, num_logprobs) -> None:
|
|
run_awq_test(
|
|
vllm_runner,
|
|
image_assets,
|
|
source_model,
|
|
quant_model,
|
|
size_factors=size_factors,
|
|
dtype=dtype,
|
|
max_tokens=max_tokens,
|
|
num_logprobs=num_logprobs,
|
|
tensor_parallel_size=1,
|
|
)
|