InternVL

LMDeploy 支持以下 InternVL 系列模型，详细信息如下表所示：

模型	大小	支持的推理引擎
InternVL	13B-19B	TurboMind
InternVL1.5	2B-26B	TurboMind, PyTorch
InternVL2	1B, 4B	PyTorch
InternVL2	2B, 8B-76B	TurboMind, PyTorch
单声道-InternVL	2B	PyTorch

下一章将演示如何使用LMDeploy部署InternVL模型，以InternVL2-8B为例。

安装

请按照安装指南安装LMDeploy，并安装InternVL2所需的其他包

pip install timm
# It is recommended to find the whl package that matches the environment from the releases on https://github.com/Dao-AILab/flash-attention.
pip install flash-attn

或者，您可以构建一个docker镜像来设置推理环境。如果您的宿主机上的CUDA版本是>=12.4，您可以运行：

docker build --build-arg CUDA_VERSION=cu12 -t openmmlab/lmdeploy:internvl . -f ./docker/InternVL_Dockerfile

否则，你可以选择：

git clone https://github.com/InternLM/lmdeploy.git
cd lmdeploy
docker build --build-arg CUDA_VERSION=cu11 -t openmmlab/lmdeploy:internvl . -f ./docker/InternVL_Dockerfile

离线推理

以下示例代码展示了VLM管道的基本用法。有关详细信息，请参阅VLM离线推理管道

from lmdeploy import pipeline
from lmdeploy.vl import load_image

pipe = pipeline('OpenGVLab/InternVL2-8B')

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

更多示例如下：

multi-image multi-round conversation, combined images

from lmdeploy import pipeline, GenerationConfig
from lmdeploy.vl.constants import IMAGE_TOKEN

pipe = pipeline('OpenGVLab/InternVL2-8B', log_level='INFO')
messages = [
    dict(role='user', content=[
        dict(type='text', text=f'{IMAGE_TOKEN}{IMAGE_TOKEN}\nDescribe the two images in detail.'),
        dict(type='image_url', image_url=dict(max_dynamic_patch=12, url='https://raw.githubusercontent.com/OpenGVLab/InternVL/main/internvl_chat/examples/image1.jpg')),
        dict(type='image_url', image_url=dict(max_dynamic_patch=12, url='https://raw.githubusercontent.com/OpenGVLab/InternVL/main/internvl_chat/examples/image2.jpg'))
    ])
]
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))

multi-image multi-round conversation, separate images

from lmdeploy import pipeline, GenerationConfig
from lmdeploy.vl.constants import IMAGE_TOKEN

pipe = pipeline('OpenGVLab/InternVL2-8B', log_level='INFO')
messages = [
    dict(role='user', content=[
        dict(type='text', text=f'Image-1: {IMAGE_TOKEN}\nImage-2: {IMAGE_TOKEN}\nDescribe the two images in detail.'),
        dict(type='image_url', image_url=dict(max_dynamic_patch=12, url='https://raw.githubusercontent.com/OpenGVLab/InternVL/main/internvl_chat/examples/image1.jpg')),
        dict(type='image_url', image_url=dict(max_dynamic_patch=12, url='https://raw.githubusercontent.com/OpenGVLab/InternVL/main/internvl_chat/examples/image2.jpg'))
    ])
]
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))

video multi-round conversation

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
pipe = pipeline('OpenGVLab/InternVL2-8B', log_level='INFO')


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


video_path = 'red-panda.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 is the red panda 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))

在线服务

你可以通过lmdeploy serve api_server CLI启动服务器：

lmdeploy serve api_server OpenGVLab/InternVL2-8B

你也可以使用上述构建的docker镜像来启动服务：

docker run --runtime nvidia --gpus all \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HUGGING_FACE_HUB_TOKEN=<secret>" \
    -p 23333:23333 \
    --ipc=host \
    openmmlab/lmdeploy:internvl \
    lmdeploy serve api_server OpenGVLab/InternVL2-8B

Docker compose 是另一种选择。在 lmdeploy 项目的根目录下创建一个 docker-compose.yml 配置文件，如下所示：

version: '3.5'

services:
  lmdeploy:
    container_name: lmdeploy
    image: openmmlab/lmdeploy:internvl
    ports:
      - "23333:23333"
    environment:
      HUGGING_FACE_HUB_TOKEN: <secret>
    volumes:
      - ~/.cache/huggingface:/root/.cache/huggingface
    stdin_open: true
    tty: true
    ipc: host
    command: lmdeploy serve api_server OpenGVLab/InternVL2-8B
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: "all"
              capabilities: [gpu]

然后，您可以执行如下启动命令：

docker-compose up -d

如果你在运行docker logs -f lmdeploy后看到以下日志，这意味着服务已成功启动。

HINT:    Please open  http://0.0.0.0:23333   in a browser for detailed api usage!!!
HINT:    Please open  http://0.0.0.0:23333   in a browser for detailed api usage!!!
HINT:    Please open  http://0.0.0.0:23333   in a browser for detailed api usage!!!
INFO:     Started server process [2439]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on  http://0.0.0.0:23333  (Press CTRL+C to quit)

lmdeploy serve api_server 的参数可以通过 lmdeploy serve api_server -h 详细查看。

有关api_server的更多信息以及如何访问该服务，可以从这里找到