开源模型应用落地-从源代码构建和运行vLLM-以满足您更高的需求
一、前言通过vLLM源码构建Docker镜像有诸多显著的好处。起首，源码构建可以或许确保我们使用的是最新的功能，制止版本不同等带来的问题。其次，自界说构建过程使可以或许根据特定需求优化镜像，好比去除不须要的依靠或者添加自界说设置，从而进步镜像的性能和安全性。此外，Docker容器化的特性使得部署更加灵活，简化了环境隔离与管理，开辟人员可以在任何支持Docker的平台上快速复现雷同的运行环境，确保开辟和生产的同等性。末了，借助Docker的版本控制和可复制性，可以更轻松地追踪和回滚到先前的镜像版本，这在举行调试和测试时尤为告急。
二、术语
2.1. Docker
是一个开源的容器化平台，允许开辟者将应用及其依靠打包成轻量级、可移植的容器。这些容器可以在任何支持 Docker 的环境中运行，从而确保应用在差别环境中的同等性。Docker 提供了简化的开辟、测试和部署流程，使得应用的交付更加高效和灵活，同时也支持微服务架构的实现。通过隔离和资源管理，Docker 使得应用的扩展和维护变得更加便捷。
2.2. hub.docker.com
是 Docker 官方的容器镜像库和共享平台，用户可以在此存储、分享和管理 Docker 镜像。它提供了一个简单易用的界面，开辟者可以通过该平台上传自己的镜像，或从中拉取公共镜像以用于项目开辟。
2.3. vLLM
vLLM是一个开源的大模型推理加速框架，通过PagedAttention高效地管理attention中缓存的张量，实现了比HuggingFace Transformers高14-24倍的吞吐量。
三、前提条件
3.1. 基础环境及前置条件
1）操纵系统：centos7
2）Tesla V100-SXM2-32GB CUDA Version: 12.2
https://i-blog.csdnimg.cn/direct/7582a2c227394ad1bdb152cb5ae73fe6.png
3.2. Docker安装
1. 更新系统，确保你的系统是最新的：
sudo yum update -y 实行结果：
Last metadata expiration check: 2:32:22 ago on Sun 06 Oct 2024 09:48:43 AM CST.
Dependencies resolved.
Nothing to do.
Complete!
2. 安装须要的依靠：
sudo yum install -y yum-utils device-mapper-persistent-data lvm2 实行结果：
Installed:
device-mapper-event-8:1.02.177-10.el8.x86_64 device-mapper-event-libs-8:1.02.177-10.el8.x86_64 device-mapper-persistent-data-0.9.0-4.el8.x86_64 libaio-0.3.112-1.el8.x86_64 lvm2-8:2.03.12-10.el8.x86_64
lvm2-libs-8:2.03.12-10.el8.x86_64
Complete!
3. 设置Docker堆栈：
sudo yum-config-manager --add-repo https://download.docker.com/linux/centos/docker-ce.repo 实行结果：
Adding repo from: https://download.docker.com/linux/centos/docker-ce.repo
4. 安装 Docker：
sudo yum install -y docker-ce docker-ce-cli containerd.io 5. 启动Docker服务，并设置为开机自启：
sudo systemctl start docker
sudo systemctl enable docker 6. 验证 Docker 安装：
sudo docker run hello-world 假如看到 "Hello from Docker!" 的消息，说明 Docker 安装成功。
7. （可选）将用户添加到Docker组：
假如你盼望以非 root 用户身份运行 Docker 命令，可以将用户添加到 Docker 组：
sudo usermod -aG docker $USER 添加后，注销并重新登录，或者运行 `newgrp docker` 使更改生效。
3.3. 下载vLLM源码
Tags · vllm-project/vllm · GitHubA high-throughput and memory-efficient inference and serving engine for LLMs - Tags · vllm-project/vllmhttps://csdnimg.cn/release/blog_editor_html/release2.3.7/ckeditor/plugins/CsdnLink/icons/icon-default.png?t=O83Ahttps://github.com/vllm-project/vllm/tags
https://i-blog.csdnimg.cn/direct/a4c9190a45354c1bab470e73b1305b40.png
(base) # git clone --branch v0.6.2 --single-branch https://github.com/vllm-project/vllm.git
Cloning into 'vllm'...
remote: Enumerating objects: 28985, done.
remote: Total 28985 (delta 0), reused 0 (delta 0), pack-reused 28985 (from 1)
Receiving objects: 100% (28985/28985), 17.32 MiB | 6.84 MiB/s, done.
Resolving deltas: 100% (21960/21960), done.
Note: switching to '7193774b1ff8603ad5bf4598e5efba0d9a39b436'.
You are in 'detached HEAD' state. You can look around, make experimental
changes and commit them, and you can discard any commits you make in this
state without impacting any branches by switching back to a branch.
If you want to create a new branch to retain commits you create, you may
do so (now or later) by using -c with the switch command. Example:
git switch -c <new-branch-name>
Or undo this operation with:
git switch -
Turn off this advice by setting config variable advice.detachedHead to false 四、技术实现
4.1. 需求
背景：使用目前最新vLLM开源镜像，无法正常运行Qwen2-VL-7B-Instruct模型，错误描述如下：
(base) # docker run --runtime nvidia --gpus all \
> -p 9000:9000 \
> --ipc=host \
> -v /data/model/qwen2-vl-7b-instruct:/qwen2-vl-7b-instruct \
> -it --rm \
> vllm/vllm-openai:latest \
> --model /qwen2-vl-7b-instruct --dtype float16 --max-parallel-loading-workers 1--max-model-len 8192 --enforce-eager --host 0.0.0.0 --port 9000
INFO 10-09 22:34:49 api_server.py:526] vLLM API server version 0.6.1.dev238+ge2c6e0a82
INFO 10-09 22:34:49 api_server.py:527] args: Namespace(host='0.0.0.0', port=9000, uvicorn_log_level='info', allow_credentials=False, allowed_origins=['*'], allowed_methods=['*'], allowed_headers=['*'], api_key=None, lora_modules=None, prompt_adapters=None, chat_template=None, response_role='assistant', ssl_keyfile=None, ssl_certfile=None, ssl_ca_certs=None, ssl_cert_reqs=0, root_path=None, middleware=[], return_tokens_as_token_ids=False, disable_frontend_multiprocessing=False, enable_auto_tool_choice=False, tool_call_parser=None, model='/qwen2-vl-7b-instruct', tokenizer=None, skip_tokenizer_init=False, revision=None, code_revision=None, tokenizer_revision=None, tokenizer_mode='auto', trust_remote_code=False, download_dir=None, load_format='auto', config_format='auto', dtype='float16', kv_cache_dtype='auto', quantization_param_path=None, max_model_len=8192, guided_decoding_backend='outlines', distributed_executor_backend=None, worker_use_ray=False, pipeline_parallel_size=1, tensor_parallel_size=1, max_parallel_loading_workers=1, ray_workers_use_nsight=False, block_size=16, enable_prefix_caching=False, disable_sliding_window=False, use_v2_block_manager=False, num_lookahead_slots=0, seed=0, swap_space=4, cpu_offload_gb=0, gpu_memory_utilization=0.9, num_gpu_blocks_override=None, max_num_batched_tokens=None, max_num_seqs=256, max_logprobs=20, disable_log_stats=False, quantization=None, rope_scaling=None, rope_theta=None, enforce_eager=True, max_context_len_to_capture=None, max_seq_len_to_capture=8192, disable_custom_all_reduce=False, tokenizer_pool_size=0, tokenizer_pool_type='ray', tokenizer_pool_extra_config=None, limit_mm_per_prompt=None, mm_processor_kwargs=None, enable_lora=False, max_loras=1, max_lora_rank=16, lora_extra_vocab_size=256, lora_dtype='auto', long_lora_scaling_factors=None, max_cpu_loras=None, fully_sharded_loras=False, enable_prompt_adapter=False, max_prompt_adapters=1, max_prompt_adapter_token=0, device='auto', num_scheduler_steps=1, multi_step_stream_outputs=False, scheduler_delay_factor=0.0, enable_chunked_prefill=None, speculative_model=None, speculative_model_quantization=None, num_speculative_tokens=None, speculative_draft_tensor_parallel_size=None, speculative_max_model_len=None, speculative_disable_by_batch_size=None, ngram_prompt_lookup_max=None, ngram_prompt_lookup_min=None, spec_decoding_acceptance_method='rejection_sampler', typical_acceptance_sampler_posterior_threshold=None, typical_acceptance_sampler_posterior_alpha=None, disable_logprobs_during_spec_decoding=None, model_loader_extra_config=None, ignore_patterns=[], preemption_mode=None, served_model_name=None, qlora_adapter_name_or_path=None, otlp_traces_endpoint=None, collect_detailed_traces=None, disable_async_output_proc=False, override_neuron_config=None, disable_log_requests=False, max_log_len=None, disable_fastapi_docs=False)
INFO 10-09 22:34:49 api_server.py:164] Multiprocessing frontend to use ipc:///tmp/11e808d4-b43b-40e9-9201-01436c4f6469 for IPC Path.
INFO 10-09 22:34:49 api_server.py:177] Started engine process with PID 22
Unrecognized keys in `rope_scaling` for 'rope_type'='default': {'mrope_section'}
Traceback (most recent call last):
File "<frozen runpy>", line 198, in _run_module_as_main
File "<frozen runpy>", line 88, in _run_code
File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 571, in <module>
uvloop.run(run_server(args))
File "/usr/local/lib/python3.12/dist-packages/uvloop/__init__.py", line 109, in run
return __asyncio.run(
^^^^^^^^^^^^^^
File "/usr/lib/python3.12/asyncio/runners.py", line 194, in run
return runner.run(main)
^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/asyncio/runners.py", line 118, in run
return self._loop.run_until_complete(task)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "uvloop/loop.pyx", line 1517, in uvloop.loop.Loop.run_until_complete
File "/usr/local/lib/python3.12/dist-packages/uvloop/__init__.py", line 61, in wrapper
return await main
^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 538, in run_server
async with build_async_engine_client(args) as engine_client:
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
return await anext(self.gen)
^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 105, in build_async_engine_client
async with build_async_engine_client_from_engine_args(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
return await anext(self.gen)
^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 182, in build_async_engine_client_from_engine_args
engine_config = engine_args.create_engine_config()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/arg_utils.py", line 874, in create_engine_config
model_config = self.create_model_config()
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/arg_utils.py", line 811, in create_model_config
return ModelConfig(
^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/config.py", line 207, in __init__
self.max_model_len = _get_and_verify_max_len(
^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/config.py", line 1746, in _get_and_verify_max_len
assert "factor" in rope_scaling
^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError
Unrecognized keys in `rope_scaling` for 'rope_type'='default': {'mrope_section'}
Process SpawnProcess-1:
Traceback (most recent call last):
File "/usr/lib/python3.12/multiprocessing/process.py", line 314, in _bootstrap
self.run()
File "/usr/lib/python3.12/multiprocessing/process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/multiprocessing/engine.py", line 388, in run_mp_engine
engine = MQLLMEngine.from_engine_args(engine_args=engine_args,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/multiprocessing/engine.py", line 134, in from_engine_args
engine_config = engine_args.create_engine_config()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/arg_utils.py", line 874, in create_engine_config
model_config = self.create_model_config()
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/arg_utils.py", line 811, in create_model_config
return ModelConfig(
^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/config.py", line 207, in __init__
self.max_model_len = _get_and_verify_max_len(
^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/config.py", line 1746, in _get_and_verify_max_len
assert "factor" in rope_scaling
^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError 目标：构建自界说vLLM镜像，可以通过docker容器快速体验Qwen2-VL-7B-Instruct效果
4.2. 添加自界说需求
4.2.1.修改transformers版本
默认vLLM源码中，指定安装的transformers版本为≥4.45.0，为了后续安装Qwen2-VL-7B-Instruct，需要将transformers指定为特定版本，示比方下：
修改requirements-common.txt文件
https://i-blog.csdnimg.cn/direct/b8e1b21a223e4d65b1ac2f7b55c0b522.png
删掉transformers依靠：
https://i-blog.csdnimg.cn/direct/aeefa034bc084fde8a5a52dffd9b740e.png
修改Dockerfile文件，指定transformers通过源码安装
RUN --mount=type=cache,target=/root/.cache/pip python3 -m pip install git+https://github.com/huggingface/transformers@21fac7abba2a37fae86106f87fcf9974fd1e3830 https://i-blog.csdnimg.cn/direct/2ba5926c2f7945078671582f6c2375b4.png
4.2.2.增加qwen-vl-utils依靠
修改requirements-common.txt文件，添加qwen-vl-utils依靠
4.2.3.问题修复
问题一：ModuleNotFoundError: No module named 'distutils'
详细描述：
832.4 from pip._internal.cli.main import main as _main# isort:skip # noqa
832.4 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
832.4 File "/usr/lib/python3/dist-packages/pip/_internal/cli/main.py", line 10, in <module>
832.4 from pip._internal.cli.autocompletion import autocomplete
832.4 File "/usr/lib/python3/dist-packages/pip/_internal/cli/autocompletion.py", line 9, in <module>
832.4 from pip._internal.cli.main_parser import create_main_parser
832.4 File "/usr/lib/python3/dist-packages/pip/_internal/cli/main_parser.py", line 7, in <module>
832.4 from pip._internal.cli import cmdoptions
832.4 File "/usr/lib/python3/dist-packages/pip/_internal/cli/cmdoptions.py", line 19, in <module>
832.4 from distutils.util import strtobool
832.4 ModuleNotFoundError: No module named 'distutils'
------
Dockerfile:148
--------------------
147 | # Install Python and other dependencies
148 | >>> RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
149 | >>> && echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections \
150 | >>> && apt-get update -y \
151 | >>> && apt-get install -y ccache software-properties-common git curl sudo vim python3-pip \
152 | >>> && apt-get install -y ffmpeg libsm6 libxext6 libgl1 \
153 | >>> && add-apt-repository ppa:deadsnakes/ppa \
154 | >>> && apt-get update -y \
155 | >>> && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv libibverbs-dev \
156 | >>> && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \
157 | >>> && update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} \
158 | >>> && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \
159 | >>> && curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} \
160 | >>> && python3 --version && python3 -m pip --version
161 |
--------------------
ERROR: failed to solve: process "/bin/sh -c echo 'tzdata tzdata/Areas select America' | debconf-set-selections && echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections && apt-get update -y && apt-get install -y ccache software-properties-common git curl sudo vim python3-pip && apt-get install -y ffmpeg libsm6 libxext6 libgl1 && add-apt-repository ppa:deadsnakes/ppa && apt-get update -y && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv libibverbs-dev && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 && update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config && curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} && python3 --version && python3 -m pip --version" did not complete successfully: exit code: 1
修改Dockerfile文件，增加python3-distutils依靠
https://i-blog.csdnimg.cn/direct/b686c37a1d7e4eed8752856b1ea37d82.png
4.3. 构建镜像
DOCKER_BUILDKIT=1 docker build . --target vllm-openai --tag vllm/vllm-qwen2-vl --build-arg max_jobs=8 --build-arg nvcc_threads=2 参数解释：
1. `DOCKER_BUILDKIT=1`:
- 这个环境变量设置了 Docker BuildKit，为 Docker 构建过程启用 BuildKit。一旦启用，BuildKit 可以提供更快的构建速度、更好的缓存机制和更强的功能。
2. `docker build`:
- 这是 Docker 命令，用于构建 Docker 镜像。
3. `.`:
- 这个点表示当前目录，Docker 会在当前目录中查找 Dockerfile 和上下文文件。上下文文件包括 Dockerfile 所需的其他文件和目录。
4. `--target vllm-openai`:
- 指定了构建的目标（stage）。在多阶段构建中，可能会有多个构建阶段。通过这个参数，Docker 会只构建名为 `vllm-openai` 的阶段。
5. `--tag vllm/vllm-qwen2-vl`:
- 指定了构建完成的镜像的名称和标签。格式是 `repository:tag`，在这里 `vllm` 是命名空间（或仓库名），`vllm-qwen2-vl` 是镜像名称。可以用于推送到 Docker Hub 或其他镜像仓库。
6. `--build-arg max_jobs=8`:
- 这个参数用于设置构建时的变量 `max_jobs`，值为 `8`。在 Dockerfile 中可以通过 `ARG max_jobs` 来访问这个变量。通常用于定制构建过程，例如并行构建的作业数。
7. `--build-arg nvcc_threads=2`:
- 这个参数类似于前一个，用于设置另一个构建时变量 `nvcc_threads`，值为 `2`。这可能用于指定与 NVIDIA CUDA 编译相关的线程数量。 漫长的构建过程：
https://i-blog.csdnimg.cn/direct/50a2145a69f241099abd903a49df81fc.png
五、附带说明
5.1.整理Docker临时文件
在 Docker 中，假如构建失败，可能会留下临时文件和中间层，这些文件可能会占用磁盘空间。
1. 整理未使用的构建缓存
docker builder prune
使用 `-f` 选项来跳过确认提示：
docker builder prune -f
2. 全面地整理 Docker，包罗未使用的容器、网络、镜像和构建缓存
docker system prune
使用 `-f` 选项来跳过确认提示：
docker system prune -f
要删除全部未使用的镜像（包罗未被任何容器使用的镜像），可以加上 `--all` 选项：
docker system prune -a -f
3. 检查磁盘使用情况，包罗镜像、容器、数据卷和构建缓存的详细信息。
docker system df
免责声明：如果侵犯了您的权益，请联系站长，我们会及时删除侵权内容，谢谢合作！更多信息从访问主页：qidao123.com:ToB企服之家，中国第一个企服评测及商务社交产业平台。
页:
[1]