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

[Model][VLM] Add Qwen2.5-Omni model support (thinker only) (#15130)

Browse Source 
Signed-off-by: fyabc <suyang.fy@alibaba-inc.com>
Signed-off-by: Roger Wang <ywang@roblox.com>
Co-authored-by: Roger Wang <136131678+ywang96@users.noreply.github.com>
Co-authored-by: Roger Wang <ywang@roblox.com>
Co-authored-by: Xiong Wang <wangxiongts@163.com>
This commit is contained in:
Yang Fan
2025-04-19 14:14:36 +08:00
committed by GitHub
parent 5c9121203c
commit 2c1bd848a6
23 changed files with 1855 additions and 85 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
examples/offline_inference/audio_language.py
View File

@ -130,6 +130,36 @@ def run_qwen2_audio(question: str, audio_count: int) -> ModelRequestData:
)
# Qwen2.5-Omni
def run_qwen2_5_omni(question: str, audio_count: int):
model_name = "Qwen/Qwen2.5-Omni-7B"
engine_args = EngineArgs(
model=model_name,
max_model_len=4096,
max_num_seqs=5,
limit_mm_per_prompt={"audio": audio_count},
)
audio_in_prompt = "".join([
"<|audio_bos|><|AUDIO|><|audio_eos|>\n" for idx in range(audio_count)
])
default_system = (
"You are Qwen, a virtual human developed by the Qwen Team, Alibaba "
"Group, capable of perceiving auditory and visual inputs, as well as "
"generating text and speech.")
prompt = (f"<|im_start|>system\n{default_system}<|im_end|>\n"
"<|im_start|>user\n"
f"{audio_in_prompt}{question}<|im_end|>\n"
"<|im_start|>assistant\n")
return ModelRequestData(
engine_args=engine_args,
prompt=prompt,
)
# Ultravox 0.5-1B
def run_ultravox(question: str, audio_count: int) -> ModelRequestData:
model_name = "fixie-ai/ultravox-v0_5-llama-3_2-1b"
@ -182,6 +212,7 @@ model_example_map = {
"minicpmo": run_minicpmo,
"phi4_mm": run_phi4mm,
"qwen2_audio": run_qwen2_audio,
"qwen2_5_omni": run_qwen2_5_omni,
"ultravox": run_ultravox,
"whisper": run_whisper,
}

32
examples/offline_inference/qwen2_5_omni/README.md Normal file
View File

@ -0,0 +1,32 @@
# Qwen2.5-Omni Offline Inference Examples
This folder provides several example scripts on how to inference Qwen2.5-Omni offline.
## Thinker Only
```bash
# Audio + image + video
python examples/offline_inference/qwen2_5_omni/only_thinker.py -q mixed_modalities
# Read vision and audio inputs from a single video file
# NOTE: V1 engine does not support interleaved modalities yet.
VLLM_USE_V1=0 python examples/offline_inference/qwen2_5_omni/only_thinker.py -q use_audio_in_video
# Multiple audios
VLLM_USE_V1=0 python examples/offline_inference/qwen2_5_omni/only_thinker.py -q multi_audios
```
This script will run the thinker part of Qwen2.5-Omni, and generate text response.
You can also test Qwen2.5-Omni on a single modality:
```bash
# Process audio inputs
python examples/offline_inference/audio_language.py --model-type qwen2_5_omni
# Process image inputs
python examples/offline_inference/vision_language.py --modality image --model-type qwen2_5_omni
# Process video inputs
python examples/offline_inference/vision_language.py --modality video --model-type qwen2_5_omni
```

160
examples/offline_inference/qwen2_5_omni/only_thinker.py Normal file
View File

@ -0,0 +1,160 @@
# SPDX-License-Identifier: Apache-2.0
"""
This example shows how to use vLLM for running offline inference
with the correct prompt format on Qwen2.5-Omni (thinker only).
"""
from typing import NamedTuple
import vllm.envs as envs
from vllm import LLM, SamplingParams
from vllm.assets.audio import AudioAsset
from vllm.assets.image import ImageAsset
from vllm.assets.video import VideoAsset
from vllm.utils import FlexibleArgumentParser
class QueryResult(NamedTuple):
inputs: dict
limit_mm_per_prompt: dict[str, int]
# NOTE: The default `max_num_seqs` and `max_model_len` may result in OOM on
# lower-end GPUs.
# Unless specified, these settings have been tested to work on a single L4.
default_system = (
"You are Qwen, a virtual human developed by the Qwen Team, Alibaba "
"Group, capable of perceiving auditory and visual inputs, as well as "
"generating text and speech.")
def get_mixed_modalities_query() -> QueryResult:
question = ("What is recited in the audio? "
"What is the content of this image? Why is this video funny?")
prompt = (f"<|im_start|>system\n{default_system}<|im_end|>\n"
"<|im_start|>user\n<|audio_bos|><|AUDIO|><|audio_eos|>"
"<|vision_bos|><|IMAGE|><|vision_eos|>"
"<|vision_bos|><|VIDEO|><|vision_eos|>"
f"{question}<|im_end|>\n"
f"<|im_start|>assistant\n")
return QueryResult(
inputs={
"prompt": prompt,
"multi_modal_data": {
"audio":
AudioAsset("mary_had_lamb").audio_and_sample_rate,
"image":
ImageAsset("cherry_blossom").pil_image.convert("RGB"),
"video":
VideoAsset(name="sample_demo_1.mp4",
num_frames=16).np_ndarrays,
},
},
limit_mm_per_prompt={
"audio": 1,
"image": 1,
"video": 1
},
)
def get_use_audio_in_video_query() -> QueryResult:
question = ("Describe the content of the video, "
"then convert what the baby say into text.")
prompt = (f"<|im_start|>system\n{default_system}<|im_end|>\n"
"<|im_start|>user\n<|vision_bos|><|VIDEO|><|vision_eos|>"
f"{question}<|im_end|>\n"
f"<|im_start|>assistant\n")
asset = VideoAsset(name="sample_demo_1.mp4", num_frames=16)
audio = asset.get_audio(sampling_rate=16000)
assert not envs.VLLM_USE_V1, ("V1 does not support use_audio_in_video. "
"Please launch this example with "
"`VLLM_USE_V1=0`.")
return QueryResult(
inputs={
"prompt": prompt,
"multi_modal_data": {
"video": asset.np_ndarrays,
"audio": audio,
},
"mm_processor_kwargs": {
"use_audio_in_video": True,
},
},
limit_mm_per_prompt={
"audio": 1,
"video": 1
},
)
def get_multi_audios_query() -> QueryResult:
question = "Are these two audio clips the same?"
prompt = (f"<|im_start|>system\n{default_system}<|im_end|>\n"
"<|im_start|>user\n<|audio_bos|><|AUDIO|><|audio_eos|>"
"<|audio_bos|><|AUDIO|><|audio_eos|>"
f"{question}<|im_end|>\n"
f"<|im_start|>assistant\n")
return QueryResult(
inputs={
"prompt": prompt,
"multi_modal_data": {
"audio": [
AudioAsset("winning_call").audio_and_sample_rate,
AudioAsset("mary_had_lamb").audio_and_sample_rate,
],
},
},
limit_mm_per_prompt={
"audio": 2,
},
)
query_map = {
"mixed_modalities": get_mixed_modalities_query,
"use_audio_in_video": get_use_audio_in_video_query,
"multi_audios": get_multi_audios_query,
}
def main(args):
model_name = "Qwen/Qwen2.5-Omni-7B"
query_result = query_map[args.query_type]()
llm = LLM(model=model_name,
max_model_len=5632,
max_num_seqs=5,
limit_mm_per_prompt=query_result.limit_mm_per_prompt,
seed=args.seed)
# We set temperature to 0.2 so that outputs can be different
# even when all prompts are identical when running batch inference.
sampling_params = SamplingParams(temperature=0.2, max_tokens=64)
outputs = llm.generate(query_result.inputs,
sampling_params=sampling_params)
for o in outputs:
generated_text = o.outputs[0].text
print(generated_text)
if __name__ == "__main__":
parser = FlexibleArgumentParser(
description='Demo on using vLLM for offline inference with '
'audio language models')
parser.add_argument('--query-type',
'-q',
type=str,
default="mixed_modalities",
choices=query_map.keys(),
help='Query type.')
parser.add_argument("--seed",
type=int,
default=None,
help="Set the seed when initializing `vllm.LLM`.")
args = parser.parse_args()
main(args)

37
examples/offline_inference/vision_language.py
View File

@ -941,6 +941,42 @@ def run_qwen2_5_vl(questions: list[str], modality: str) -> ModelRequestData:
)
# Qwen2.5-Omni
def run_qwen2_5_omni(questions: list[str], modality: str):
model_name = "Qwen/Qwen2.5-Omni-7B"
engine_args = EngineArgs(
model=model_name,
max_model_len=4096,
max_num_seqs=5,
mm_processor_kwargs={
"min_pixels": 28 * 28,
"max_pixels": 1280 * 28 * 28,
"fps": [1],
},
disable_mm_preprocessor_cache=args.disable_mm_preprocessor_cache,
)
if modality == "image":
placeholder = "<|IMAGE|>"
elif modality == "video":
placeholder = "<|VIDEO|>"
default_system = (
"You are Qwen, a virtual human developed by the Qwen Team, Alibaba "
"Group, capable of perceiving auditory and visual inputs, as well as "
"generating text and speech.")
prompts = [(f"<|im_start|>system\n{default_system}<|im_end|>\n"
f"<|im_start|>user\n<|vision_bos|>{placeholder}<|vision_eos|>"
f"{question}<|im_end|>\n"
"<|im_start|>assistant\n") for question in questions]
return ModelRequestData(
engine_args=engine_args,
prompts=prompts,
)
# SkyworkR1V
def run_skyworkr1v(questions: list[str], modality: str) -> ModelRequestData:
assert modality == "image"
@ -1010,6 +1046,7 @@ model_example_map = {
"qwen_vl": run_qwen_vl,
"qwen2_vl": run_qwen2_vl,
"qwen2_5_vl": run_qwen2_5_vl,
"qwen2_5_omni": run_qwen2_5_omni,
"skywork_chat": run_skyworkr1v,
"smolvlm": run_smolvlm,
}