[Feature] support qwen2.5-vl for pytorch engine

[CUHKSZzxy]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily receiving feedbacks. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

Issue #3132, support Qwen2.5-vl 3/7/72 B

Modification

1. Include Qwen2.5-VL modeling
2. Include documents

BC-breaking (Optional)

Does the modification introduce changes that break the backward-compatibility of the downstream repositories?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

1. Single-round conversation:

import os
from lmdeploy import pipeline
from lmdeploy.vl import load_image


os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-3B-Instruct/snapshots/1b989f2c63999d7344135894d3cfa8f494116743'
model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19'
# model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-72B-Instruct/snapshots/5d8e171e5ee60e8ca4c6daa380bd29f78fe19021'

pipe = pipeline(model_path)

image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')
response = pipe((f'describe this image', image))
print('-'*50)
print(response)

2. Multi-round conversations

import os
from lmdeploy import pipeline, GenerationConfig

os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-3B-Instruct/snapshots/1b989f2c63999d7344135894d3cfa8f494116743'
model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19'
# model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-72B-Instruct/snapshots/5d8e171e5ee60e8ca4c6daa380bd29f78fe19021'

pipe = pipeline(model_path)

messages = [
    dict(role='user', content=[
        dict(type='text', text='Describe the two images in detail.'),
        dict(type='image_url', image_url=dict(url='https://raw.githubusercontent.com/QwenLM/Qwen-VL/master/assets/mm_tutorial/Beijing_Small.jpeg')),
        dict(type='image_url', image_url=dict(url='https://raw.githubusercontent.com/QwenLM/Qwen-VL/master/assets/mm_tutorial/Chongqing_Small.jpeg'))
    ])
]
out = pipe(messages, gen_config=GenerationConfig(top_k=1))

messages.append(dict(role='assistant', content=out.text))
messages.append(dict(role='user', content='What are the similarities and differences between these two images.'))
out = pipe(messages, gen_config=GenerationConfig(top_k=1))
print(f'-'*50)
print(out)

3. Boost performance

import os
from lmdeploy import pipeline, GenerationConfig

os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-3B-Instruct/snapshots/1b989f2c63999d7344135894d3cfa8f494116743'
model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19'
# model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-72B-Instruct/snapshots/5d8e171e5ee60e8ca4c6daa380bd29f78fe19021'

pipe = pipeline(model_path)

min_pixels = 64 * 28 * 28
max_pixels = 64 * 28 * 28
messages = [
    dict(role='user', content=[
        dict(type='text', text='Describe the two images in detail.'),
        dict(type='image_url', image_url=dict(min_pixels=min_pixels, max_pixels=max_pixels, url='https://raw.githubusercontent.com/QwenLM/Qwen-VL/master/assets/mm_tutorial/Beijing_Small.jpeg')),
        dict(type='image_url', image_url=dict(min_pixels=min_pixels, max_pixels=max_pixels, url='https://raw.githubusercontent.com/QwenLM/Qwen-VL/master/assets/mm_tutorial/Chongqing_Small.jpeg'))
    ])
]
out = pipe(messages, gen_config=GenerationConfig(top_k=1))

messages.append(dict(role='assistant', content=out.text))
messages.append(dict(role='user', content='What are the similarities and differences between these two images.'))
out = pipe(messages, gen_config=GenerationConfig(top_k=1))
print(f'-'*50)
print(out)

4. Multi-round video understanding

import os
import numpy as np
from lmdeploy import pipeline, GenerationConfig
from decord import VideoReader, cpu
from lmdeploy.vl.constants import IMAGE_TOKEN
from lmdeploy.vl.utils import encode_image_base64
from PIL import Image

os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-3B-Instruct/snapshots/1b989f2c63999d7344135894d3cfa8f494116743'
model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19'
# model_path = '/mnt/141/huggingface_hub/hub/models--Qwen--Qwen2.5-VL-72B-Instruct/snapshots/5d8e171e5ee60e8ca4c6daa380bd29f78fe19021'

pipe = pipeline(model_path)


def get_index(bound, fps, max_frame, first_idx=0, num_segments=32):
    if bound:
        start, end = bound[0], bound[1]
    else:
        start, end = -100000, 100000
    start_idx = max(first_idx, round(start * fps))
    end_idx = min(round(end * fps), max_frame)
    seg_size = float(end_idx - start_idx) / num_segments
    frame_indices = np.array([
        int(start_idx + (seg_size / 2) + np.round(seg_size * idx))
        for idx in range(num_segments)
    ])
    return frame_indices


def load_video(video_path, bound=None, num_segments=32):
    vr = VideoReader(video_path, ctx=cpu(0), num_threads=1)
    max_frame = len(vr) - 1
    fps = float(vr.get_avg_fps())
    pixel_values_list, num_patches_list = [], []
    frame_indices = get_index(bound, fps, max_frame, first_idx=0, num_segments=num_segments)
    imgs = []
    for frame_index in frame_indices:
        img = Image.fromarray(vr[frame_index].asnumpy()).convert('RGB')
        imgs.append(img)
    return imgs


# NOTE: download video from: https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/space_woaudio.mp4
video_path = '/home/zhouxinyu1/lmdeploy_qwen/0_test/space_woaudio.mp4'
imgs = load_video(video_path, num_segments=8)

question = ''
for i in range(len(imgs)):
    question = question + f'Frame{i+1}: {IMAGE_TOKEN}\n'

question += 'What are they doing?'

content = [{'type': 'text', 'text': question}]
for img in imgs:
    content.append({'type': 'image_url', 'image_url': {'max_dynamic_patch': 1, 'url': f'data:image/jpeg;base64,{encode_image_base64(img)}'}})

messages = [dict(role='user', content=content)]
out = pipe(messages, gen_config=GenerationConfig(top_k=1))

messages.append(dict(role='assistant', content=out.text))
messages.append(dict(role='user', content='Describe this video in detail. Don\'t repeat.'))
out = pipe(messages, gen_config=GenerationConfig(top_k=1))
print(f'-'*50)
print(out)

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit tests to ensure the correctness.
3. If the modification has a dependency on downstream projects of a newer version, this PR should be tested with all supported versions of downstream projects.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[lvhan028]

cc @zhulinJulia24 may add qwen2.5-vl model to test cases

[irexyc]

Since we only support image input, I think we can reuse vl/model/qwen2.py.

Changing this line to _arch = ['Qwen2VLForConditionalGeneration', 'Qwen2_5_VLForConditionalGeneration'] should be enough for pytorch inference.

[RunningLeon]

LGTM

[lvhan028]

@zhulinJulia24 may add qwen2.5-vl into test set