InternVL2 Docker 镜像使⽤说明
俗世奇人 by InternVL2 Docker 镜像使⽤说明 概述 本⽂旨在为⽤⼾提供⼀个⽅便的⽅式来配置所需的 Python 环境，以便运⾏InternVL2。通过使⽤ Docker，⽤⼾可以在隔离的容器中轻松设置和管理依赖项，⽆需担⼼与系统环境的冲突或其他依赖问 题。本⽂以运⾏InternVL2-1B为⽰例给出详细步骤，如有遇到报错请先参考： https://docs.docker.com/manuals/ https://yeasy.gitbook.io/docker_practice https://chatgpt.com/ ⽅法1:在线拉取 前提条件 1. 安装了 Docker 的系统 2. 可以通过某种⽅式顺畅连接到 https://hub.docker.com/ 3. cuda>=11.8 步骤 1 ：拉取Docker镜像 1. 使⽤ docker pull 命令拉取镜像。命令格式为： 1 docker pull qishisurenhhh/internvl:latest 如果你没有指定 <tag> ，默认会拉取 latest 标签的镜像。 1 latest: Pulling from qishisurenhhh/internvl 2 Digest: sha256: 68 db47419cb037d27e3afd6d978fb78d630e4bc0d2d33096350189a337ad9fa5 3 Status: Image is up to date for qishisurenhhh/internvl:latest 4 docker.io/qishisurenhhh/internvl:latest 步骤 2：运⾏ Docker 容器 1. 运⾏ Docker 容器： 使⽤以下命令运⾏加载的 Docker 镜像，并根据需要覆盖默认环境变量： 1 docker run --gpus all \ 2 -e CUDA_DEVICE=0 \ 3 -e MODEL_PATH= 'OpenGVLab/InternVL2-1B' \ 4 -e http_proxy= \ 5 -e https_proxy= \ 6 -e HTTP_PROXY= \ 7 -e HTTPS_PROXY= \ 8 -it qishisuren/internvl:v2 docker run 这是⽤于运⾏容器的 Docker 命令。 --gpus all 这部分⽤于告诉 Docker 运⾏的容器可以访问所有可⽤的 GPU。确保 Docker 守护进程配置了 NVIDIA GPU ⽀持。 -e CUDA_DEVICE=0 这⾏⽤于设置容器内的环境变量 CUDA_DEVICE ，值为 0 。这个变量可以被容器内的应⽤程序访 问，⽤于指定要使⽤的 CUDA 设备编号。 -e MODEL_PATH='path/to/other/model' 这⾏⽤于设置容器内的环境变量 MODEL_PATH ，默认值为 'OpenGVLab/InternVL2-1B' 。 -e http_proxy= -e https_proxy= -e HTTP_PROXY= -e HTTPS_PROXY= 如果您的容器需要通过代理访问互联⽹，可以在此设置代理地址。 -it 这部分表⽰以交互模式运⾏容器，并且分配⼀个伪终端。它组合了 -i （交互式）和 -t （伪终端） 两个选项。 输出 1 ========== 2 == CUDA == 3 ========== 4 5 CUDA Version 11.8.0 6 7 Container image Copyright (c) 2016-2023 , NVIDIA CORPORATION & AFFILIATES. All rights reserved. 8 9 This container image and its contents are governed by the NVIDIA Deep Learning Container License. 10 By pulling and using the container, you accept the terms and conditions of this license: 11 https: //developer.nvidia.com/ngc/nvidia-deep-learning-container-license 12 13 A copy of this license is made available in this container at /NGC-DL CONTAINER-LICENSE for your convenience. 14 15 Using CUDA device: 4 16 Using model path: OpenGVLab/InternVL2 -1 B 17 /usr/local/lib/python3 .10 /dist-packages/huggingface_hub/file_download.py: 1150 : FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0 . Downloads always resume when possible. If you want to force a new download, use `force_download=True`. 18 warnings.warn( 19 config.json: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ██████████████████████████████████████████████████████████ | 3.68 k/ 3.68 k [ 00 : 00 < 00 : 00 , 35.2 MB/s] 20 configuration_internvl_chat.py: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ███████████████████████████████████████ | 3.80 k/ 3.80 k [ 00 : 00 < 00 : 00 , 30.3 MB/s] 21 configuration_intern_vit.py: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ██████████████████████████████████████████ | 5.55 k/ 5.55 k [ 00 : 00 < 00 : 00 , 39.5 MB/s] 22 A new version of the following files was downloaded from https: //huggingface.co/OpenGVLab/InternVL2-1B: 23 - configuration_intern_vit.py 24 . Make sure to double-check they do not contain any added malicious code. To avoid downloading new versions of the code file, you can pin a revision. 25 A new version of the following files was downloaded from https: //huggingface.co/OpenGVLab/InternVL2-1B: 26 - configuration_internvl_chat.py 27 - configuration_intern_vit.py 28 . Make sure to double-check they do not contain any added malicious code. To avoid downloading new versions of the code file, you can pin a revision. 29 modeling_internvl_chat.py: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████ | 15.2 k/ 15.2 k [ 00 : 00 < 00 : 00 , 3.82 MB/s] 30 modeling_intern_vit.py: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████ | 18.2 k/ 18.2 k [ 00 : 00 < 00 : 00 , 112 MB/s] 31 A new version of the following files was downloaded from https: //huggingface.co/OpenGVLab/InternVL2-1B: 32 - modeling_intern_vit.py 33 . Make sure to double-check they do not contain any added malicious code. To avoid downloading new versions of the code file, you can pin a revision. 34 conversation.py: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ██████████████████████████████████████████████████████ | 15.0 k/ 15.0 k [ 00 : 00 < 00 : 00 , 3.99 MB/s] 35 A new version of the following files was downloaded from https: //huggingface.co/OpenGVLab/InternVL2-1B: 36 - conversation.py 37 . Make sure to double-check they do not contain any added malicious code. To avoid downloading new versions of the code file, you can pin a revision. 38 A new version of the following files was downloaded from https: //huggingface.co/OpenGVLab/InternVL2-1B: 39 - modeling_internvl_chat.py 40 - modeling_intern_vit.py 41 - conversation.py 42 . Make sure to double-check they do not contain any added malicious code. To avoid downloading new versions of the code file, you can pin a revision. 43 /usr/local/lib/python3 .10 /dist-packages/huggingface_hub/file_download.py: 1150 : FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0 . Downloads always resume when possible. If you want to force a new download, use `force_download=True`. 44 warnings.warn( 45 model.safetensors: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████ | 1.88 G/ 1.88 G [ 03 : 00 < 00 : 00 , 10.4 MB/s] 46 generation_config.json: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ██████████████████████████████████████████████████ | 69.0 / 69.0 [ 00 : 00 < 00 : 00 , 722 kB/s] 47 tokenizer_config.json: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████ | 3.02 k/ 3.02 k [ 00 : 00 < 00 : 00 , 34.3 MB/s] 48 vocab.json: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ███████████████████████████████████████████████████████████ | 3.38 M/ 3.38 M [ 00 : 00 < 00 : 00 , 3.49 MB/s] 49 merges.txt: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ███████████████████████████████████████████████████████████ | 1.67 M/ 1.67 M [ 00 : 00 < 00 : 00 , 2.77 MB/s] 50 added_tokens.json: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████ | 265 / 265 [ 00 : 00 < 00 : 00 , 3.35 MB/s] 51 special_tokens_map.json: 100 %| ████████████████████████████████████████████████████████████████ ████████████████████████████████████████████████████████████████ ██████████████████████████████████████████████████ | 498 / 498 [ 00 : 00 < 00 : 00 , 6.76 MB/s] 52 Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained. 53 Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained. 54 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 55 User: Hello, who are you? 56 Assistant: I am an AI assistant whose name is InternVL, developed jointly by Shanghai AI Lab and SenseTime. 57 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 58 User: Can you tell me a story? 59 Assistant: Sure, here's a story: 60 61 Once upon a time, there was a young girl named Lily who lived in a small village. She was a curious and adventurous girl who loved exploring the world around her. One day, while wandering through the forest, she stumbled upon a hidden cave. Excited by the prospect of discovering something new, she decided to explore deeper into the cave. 62 63 As she entered the cave, she was amazed by the beauty of the surroundings. The cave was filled with colorful flowers, sparkling crystals, and strange creatures. Lily felt a sense of wonder and excitement as she explored the cave further. 64 65 Suddenly, she heard a loud noise coming from the cave. She quickly turned around and saw a group of angry creatures guarding the entrance. Lily was scared, but she knew that she had to stay calm and find a way out. 66 67 She used her wits and skills to outsmart the creatures and escape the cave. She made her way through the cave, dodging obstacles and avoiding traps. Finally, she reached the exit and emerged from the cave safely. 68 69 As she looked back at the cave, Lily realized that she had discovered something truly magical. She had discovered a world of wonder and adventure, and she knew that she would always be fascinated by the mysteries of the world around her. 70 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 71 User: <image> 72 Please describe the image shortly. 73 Assistant: The image shows a red panda, a small, furry mammal known for its distinctive red and white fur. The red panda is resting on a wooden structure, possibly a platform or a bench, with its head leaning over the edge. The animal's large, expressive eyes are looking directly at the camera, giving it a curious and alert appearance. The background features a blurred natural setting with greenery, suggesting that the photo was taken in a zoo or a wildlife park. 74 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 75 User: <image> 76 Please describe the image in detail. 77 Assistant: The image depicts a red panda, a small, intelligent mammal known for its distinctive red and white fur. The red panda is characterized by its large ears, which are covered in fur, and its large, expressive eyes. The animal is resting on a wooden structure, which appears to be part of a wooden enclosure, possibly a zoo or wildlife park. The background is blurred, but it shows greenery, indicating that the setting is outdoors. The red panda's fur is predominantly red, with white patches on its face, ears, and limbs. The image captures the red panda in a relaxed and curious pose, with its eyes looking directly at the camera. 78 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 79 User: Please write a poem according to the image. 80 Assistant: In the forest's gentle embrace, 81 A red panda rests, serene and still, 82 Its fur a vivid red, its face a blaze, 83 A sight to behold, a marvel to see. 84 85 Its ears, large and round, 86 Covered with fur, a sight to see, 87 A blend of colors, a marvel to see, 88 A sight to behold, a marvel to see. 89 90 The background, blurred green, 91 Nature's canvas, a sight to see, 92 A peaceful sanctuary, a sight to see, 93 A sanctuary of life, a sight to see. 94 95 The red panda's eyes, 96 A soulful gaze, a sight to see, 97 A gentle presence, a sight to see, 98 A sight to behold, a sight to see. 99 100 In the forest's gentle embrace, 101 A red panda rests, serene and still, 102 A sight to behold, a marvel to see, 103 A sight to behold, a marvel to see. 104 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 105 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 106 User: What are the similarities and differences between these two images. 107 Assistant: The two images depict two different animals, but they share some similarities and differences. Here are the key similarities and differences: 108 109 ### Similarities: 110 1. **Species**: Both images feature a panda cub. 111 2. **Location**: Both images are taken in a natural habitat, likely a zoo or wildlife park. 112 3. **Setting**: Both images show the panda resting on a wooden structure, which is part of a larger enclosure. 113 4. **Environment**: Both images have lush greenery, indicating a natural habitat environment. 114 115 ### Differences: 116 1. **Cub's Appearance**: 117 - **Cub in the first image**: The cub has a predominantly black and white fur pattern, with a white face and black patches around the eyes and mouth. 118 - **Cub in the second image**: The cub has a similar appearance, with a black and white fur pattern, but the background is more blurred, and the setting appears to be a different environment. 119 120 2. **Setting**: 121 - **First Image**: The cub is resting on a wooden structure, possibly part of a larger enclosure. 122 - **Second Image**: The cub is in a more natural, less structured environment, surrounded by greenery. 123 124 3. **Behavior**: 125 - **First Image**: The cub appears to be resting or resting, with its eyes looking directly at the camera. 126 - **Second Image**: The cub seems to be more relaxed, possibly resting or exploring its surroundings. 127 128 4. **Background**: 129 - **First Image**: The background is more natural and blurred, with greenery and trees. 130 - **Second Image**: The background is more blurred, but it still shows greenery, indicating a similar natural habitat. 131 132 These similarities and differences highlight the unique characteristics of each panda cub, while also showcasing the natural habitat and environment in which they are found. 133 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 134 User: Image -1 : <image> 135 Image -2 : <image> 136 Describe the two images in detail. 137 Assistant: The two images provided are of pandas, a species of mammal native to Asia. Both images depict pandas in their natural habitat, which appears to be a zoo or a wildlife sanctuary. 138 139 ### Image -1 : 140 - **Panda**: The image features a young panda cub. The cub has a predominantly white face with a black patch on its chest and a brownish-orange patch on its back. The cub's eyes are dark and alert, and it has a small, round nose. The cub is resting on a wooden structure, possibly a platform or a log, which is part of a larger enclosure. The background shows greenery, indicating a natural setting, likely a forest or a park. 141 142 ### Image -2 : 143 - **Panda**: The second image shows a larger panda cub, which appears to be a juvenile. This panda cub is also predominantly white with black patches on its face and back. It has a similar round nose and dark eyes. The cub is nestled among green plants, suggesting it is in a more controlled environment, possibly a zoo or a sanctuary. The background includes more greenery, indicating a dense forest or a similar natural habitat. 144 145 Both images capture the adorable and endearing nature of pandas, highlighting their social behavior and the natural environment they thrive in. The presence of greenery and the natural habitat in both images emphasizes the importance of preserving and protecting these endangered species. 146 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 147 User: What are the similarities and differences between these two images. 148 Assistant: The two images of pandas provide a fascinating comparison of their physical characteristics and the environments they are in. Here are the key similarities and differences: 149 150 ### Similarities: 151 1. **Species**: Both images depict pandas, a species of Asian rodent. 152 2. **Physical Features**: Both pandas have a white face with black patches on their back and chest, a characteristic feature of the species. 153 3. **Cub Size**: The cub in both images is relatively small compared to the adult panda, which is typical for juvenile pandas. 154 4. **Habitat**: Both images show pandas in natural settings, likely in zoos or wildlife sanctuaries. 155 156 ### Differences: 157 1. **Size and Build**: 158 - The cub in Image -1 appears to be slightly larger than the juvenile in Image -2. 159 - The adult panda in Image -1 has a more robust build, with a larger head and a more pronounced facial structure. 160 2. **Environment**: 161 - The cub in Image -1 is resting on a wooden structure, suggesting a more controlled and human-influenced environment. 162 - The cub in Image -2 is nestled among green plants, indicating a more natural and possibly more controlled habitat. 163 3. **Behavioral Traits**: 164 - The cub in Image -1 seems to be more alert and possibly more active, as it is resting. 165 - The cub in Image -2 appears more relaxed and comfortable, possibly indicating a more natural or less controlled environment. 166 4. **Cub Behavior**: 167 - The cub in Image -1 seems to be more curious and possibly more active, as it is resting. 168 - The cub in Image -2 appears more docile and possibly more comfortable in its natural habitat. 169 170 ### Conclusion: 171 - The primary similarity between the two images is the presence of pandas, a species that is native to Asia. 172 - The differences lie in the size, build, and the environments they are in. The cub in Image -1 appears to be larger and more human-influenced, while the cub in Image -2 seems to be in a more natural, possibly controlled habitat. 173 Setting `pad_token_id` to `eos_token_id`: 151645 for open-end generation. 174 User: <image> 175 Describe the image in detail. 176 Assistant: The image features a young red panda, a popular and beloved species of animal. The red panda has a striking appearance with a predominantly red face, a white underbelly, and a black nose. Its fur is thick and fluffy, with a mix of red and white fur, particularly around the face, ears, and limbs. The red panda's eyes are large and expressive, with a black pupil and a white ring around the iris, giving it a distinctive and endearing look. 177 178 The red panda is resting on a wooden structure, which appears to be part of a wooden enclosure or cage. The background is blurred but shows greenery, indicating that the red panda is likely in a natural or semi-natural habitat. The lighting in the image is soft and natural, suggesting that the photo was taken outdoors in daylight. The red panda's posture is relaxed, with its head resting on the wooden structure, and it appears to be looking directly at the camera with a curious and gentle expression. 179 User: <image> 180 Describe the image in detail. 181 Assistant: The image depicts a cute, black-and-white panda cub sitting amidst lush greenery. The panda is nestled among bamboo stalks, which are a common habitat for pandas. The cub appears to be in a relaxed state, with its front paws resting on the bamboo stalks and its head slightly tilted upwards. The panda's large, expressive eyes are wide open, and it seems to be looking directly at the camera with a curious and gentle expression. 182 183 The background is filled with various green plants and foliage, creating a natural and serene environment. The ground is covered with a mix of grass and small plants, which adds to the overall tranquil atmosphere of the scene. The lighting in the image suggests it is daytime, with natural light illuminating the scene and highlighting the vibrant colors of the panda and its surroundings. ⽅法2:离线⽅式获得 前提条件 1. 安装了 Docker 的系统。 2. 可以通过 USB 驱动器、⽹络共享或其他⽅法传输⽂件的⼿段。 3. cuda>=11.8 步骤 1：在本地机器上加载 Docker 镜像 1. 确保 Docker 已安装： 在终端中输⼊以下命令以确认 Docker 已安装： 1 docker --version 2. 将镜像压缩包 internvl_v2.tar 拷⻉到本地机器： 3. 加载 Docker 镜像： 在终端中运⾏以下命令将镜像加载到 Docker 中,此命令会从 internvl_v2 .tar ⽂件中加载 Docker 镜像，并将其添加到本地 Docker 镜像库中： 1 docker load -i internvl_v2 .tar 4. 确认镜像已加载： 使⽤以下命令列出所有本地 Docker 镜像，以确认镜像已成功加载： 1 docker images 您应该能够看到⼀个新的镜像，例如： 1 REPOSITORY TAG IMAGE ID CREATED SIZE 2 qishisuren/internvl v2 e32384490a23 2 days ago 16.6GB 步骤 2：运⾏Docker容器 同⽅法1 输出 同⽅法1 ⽅法3:⾃主构建镜像 前提条件 1. 安装了 Docker 的系统。 2. 可以通过 USB 驱动器、⽹络共享或其他⽅法传输⽂件的⼿段。 3. cuda>=11.8 4. 已复制项⽬⽂件到本地： 1 git clone https: //github.com/OpenGVLab/InternVL.git Dockerfile 1 FROM nvidia/cuda: 11.8.0 -cudnn8-devel-ubuntu22 .04 2 3 WORKDIR /InternVL 4 5 # 设置代理环境变量（如有必要） 6 ENV http_proxy= \ 7 https_proxy= \ 8 HTTP_PROXY= \ 9 HTTPS_PROXY 10 11 # 安装系统依赖项和 Python 包 12 RUN apt-get update && apt-get install -y \ 13 git \ 14 wget \ 15 python3-pip \ 16 && apt-get clean \ 17 && rm -rf /var/lib/apt/lists /* 18 19 # 设置 CUDA_HOME 环境变量 20 ENV CUDA_HOME=/usr/local/cuda 21 ENV PATH=$CUDA_HOME/bin:$PATH 22 ENV LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH 23 24 # 复制应⽤程序代码到⼯作⽬录 25 COPY . . 26 # 或从 GitHub 克隆指定仓库 27 # RUN git clone https://github.com/OpenGVLab/InternVL.git 28 29 # 进⼊仓库⽬录 30 WORKDIR /InternVL 31 32 # 安装 Python 包 33 # 注意：⼿动删除 internvl_chat.txt 中 flash-attn 字段 34 RUN pip3 install --no-cache-dir packaging \ 35 && pip3 install --no-cache-dir -r ./requirements/internvl_chat.txt \ 36 && pip3 install -U flash-attn --no-build-isolation 37 38 # 设置默认环境变量 39 ENV CUDA_DEVICE=4 40 ENV MODEL_PATH='OpenGVLab/InternVL2-1B' 41 42 # 进⼊到 /internvl_chat ⽂件夹 43 WORKDIR ./internvl_chat 44 45 # 运⾏你的应⽤程序 46 CMD ["python3", "../run.py"] 步骤 1：构建 Docker 镜像 ⾸先，在有互联⽹连接的机器上构建 Docker 镜像。 1. 确保 Dockerfile 在当前⽬录： 1 ls 2 Dockerfile 2. 构建 Docker 镜像： 1 docker build -t yourusername/internvl:latest . 步骤 2：运⾏Docker容器 同⽅法1 输出 同⽅法1
