使用 vLLM 运行 GLM-4-9B

游戏AI开发者

已于 2024-06-06 08:11:40 修改

阅读量2.8k

点赞数 26

分类专栏： LLM 文章标签：语言模型 llama gpt ai pytorch

于 2024-06-06 00:02:54 首次发布

本文链接：https://blog.csdn.net/word_world/article/details/139482777

版权

LLM 专栏收录该内容

2 篇文章 0 订阅

订阅专栏

安装 vLLM

注意：vLLM 只支持 linux 或 wsl

执行 $ pip install vllm 安装 vllm 可能遇到下面报错。

AssertionError: vLLM only supports Linux platform (including WSL).

vLLM 不能直接跑在 windows 上，OMG.

进入 wsl（如 ubuntu 22.04，python 3.10）执行：

安装 cuda

# apt-get install nvidia-cuda-toolkit
# nvcc -V

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Thu_Nov_18_09:45:30_PST_2021
Cuda compilation tools, release 11.5, V11.5.119
Build cuda_11.5.r11.5/compiler.30672275_0

安装 vllm
# pip install vllm 报错：

 ERROR: Could not find a version that satisfies the requirement vllm (from versions: none)
 ERROR: No matching distribution found for vllm

升级 pip：
# python3 -m pip install --upgrade pip

再尝试：
# pip install vllm

...
Downloading vllm-0.4.3-cp310-cp310-manylinux1_x86_64.whl (131.1 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 131.1/131.1 MB 640.5 kB/s eta 0:00:00
Downloading lm_format_enforcer-0.10.1-py3-none-any.whl (42 kB)
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 42.9/42.9 kB 569.5 kB/s eta 0:00:00
Downloading outlines-0.0.34-py3-none-any.whl (76 kB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 76.5/76.5 kB 613.2 kB/s eta 0:00:00
Downloading torch-2.3.0-cp310-cp310-manylinux1_x86_64.whl (779.1 MB)
   ━━━━━━━━━━━━━━━━━━━━╺━━━━━━━━━━━━━━━━━━━ 391.0/779.1 MB 3  65.6 kB/s eta 0:17:42
...

Downloading openai-1.31.1-py3-none-any.whl 遇到 pip._vendor.urllib3.exceptions.ReadTimeoutError

# pip install urllib3==1.25.11

手动安装 openai

# pip install openai

更新回最新 urllib3，否则其他依赖也会下载超时

# pip install --upgrade urllib3

vLLM 安装成功

再尝试：
# pip install vllm

Installing collected packages: sentencepiece, py-cpuinfo, nvidia-ml-py, ninja, mpmath, websockets, uvloop, uvicorn, ujson, sympy, shellingham, safetensors, rpds-py, regex, python-multipart, python-dotenv, pygments, psutil, protobuf, prometheus-client, packaging, orjson, nvidia-nvtx-cu12, nvidia-nvjitlink-cu12, nvidia-nccl-cu12, nvidia-curand-cu12, nvidia-cufft-cu12, nvidia-cuda-runtime-cu12, nvidia-cuda-nvrtc-cu12, nvidia-cuda-cupti-cu12, nvidia-cublas-cu12, numpy, networkx, nest-asyncio, multidict, msgpack, mdurl, llvmlite, lark, joblib, interegular, httptools, fsspec, frozenlist, filelock, dnspython, diskcache, cmake, cloudpickle, charset-normalizer, attrs, async-timeout, yarl, watchfiles, triton, starlette, scipy, requests, referencing, nvidia-cusparse-cu12, nvidia-cudnn-cu12, numba, markdown-it-py, email_validator, aiosignal, tiktoken, rich, ray, prometheus-fastapi-instrumentator, nvidia-cusolver-cu12, lm-format-enforcer, huggingface-hub, aiohttp, typer, torch, tokenizers, xformers, vllm-flash-attn, transformers, fastapi-cli, outlines, fastapi, vllm
Successfully installed aiohttp-3.9.5 aiosignal-1.3.1 async-timeout-4.0.3 attrs-23.2.0 charset-normalizer-3.3.2 cloudpickle-3.0.0 cmake-3.29.3 diskcache-5.6.3 dnspython-2.6.1 email_validator-2.1.1 fastapi-0.111.0 fastapi-cli-0.0.4 filelock-3.14.0 frozenlist-1.4.1 fsspec-2024.6.0 httptools-0.6.1 huggingface-hub-0.23.3 interegular-0.3.3 joblib-1.4.2 lark-1.1.9 llvmlite-0.42.0 lm-format-enforcer-0.10.1 markdown-it-py-3.0.0 mdurl-0.1.2 mpmath-1.3.0 msgpack-1.0.8 multidict-6.0.5 nest-asyncio-1.6.0 networkx-3.3 ninja-1.11.1.1 numba-0.59.1 numpy-1.26.4 nvidia-cublas-cu12-12.1.3.1 nvidia-cuda-cupti-cu12-12.1.105 nvidia-cuda-nvrtc-cu12-12.1.105 nvidia-cuda-runtime-cu12-12.1.105 nvidia-cudnn-cu12-8.9.2.26 nvidia-cufft-cu12-11.0.2.54 nvidia-curand-cu12-10.3.2.106 nvidia-cusolver-cu12-11.4.5.107 nvidia-cusparse-cu12-12.1.0.106 nvidia-ml-py-12.555.43 nvidia-nccl-cu12-2.20.5 nvidia-nvjitlink-cu12-12.5.40 nvidia-nvtx-cu12-12.1.105 orjson-3.10.3 outlines-0.0.34 packaging-24.0 prometheus-client-0.20.0 prometheus-fastapi-instrumentator-7.0.0 protobuf-5.27.0 psutil-5.9.8 py-cpuinfo-9.0.0 pygments-2.18.0 python-dotenv-1.0.1 python-multipart-0.0.9 ray-2.23.0 referencing-0.35.1 regex-2024.5.15 requests-2.32.3 rich-13.7.1 rpds-py-0.18.1 safetensors-0.4.3 scipy-1.13.1 sentencepiece-0.2.0 shellingham-1.5.4 starlette-0.37.2 sympy-1.12.1 tiktoken-0.7.0 tokenizers-0.19.1 torch-2.3.0 transformers-4.41.2 triton-2.3.0 typer-0.12.3 ujson-5.10.0 uvicorn-0.30.1 uvloop-0.19.0 vllm-0.4.3 vllm-flash-attn-2.5.8.post2 watchfiles-0.22.0 websockets-12.0 xformers-0.0.26.post1 yarl-1.9.4

查看 vLLM 版本

# pip list | grep vllm

vllm                              0.4.3
vllm-flash-attn                   2.5.8.post2

windows 上找不到 torch\lib\shm.dll

OSError: [WinError 126] 找不到指定的模块。 Error loading "AppData\Local\Temp\pip-build-env-aq9ya1hl\overlay\Lib\site-packages\torch\lib\shm.dll" or one of its dependencies.

参考 https://discuss.pytorch.org/t/error-loading-lib-site-packages-torch-lib-shm-dll-or-one-of-its-dependencie/201695 中的讨论。

执行 $ nvcc --version 查看本机 cuda 版本：

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Tue_Feb_27_16:28:36_Pacific_Standard_Time_2024
Cuda compilation tools, release 12.4, V12.4.99
Build cuda_12.4.r12.4/compiler.33961263_0

根据 https://pytorch.org/get-started/locally/ 的提示，由于在当前时间（2024-06-05）我的 cuda 12.4 版本过新，需要安装 Preview(Nightly) 版本的 PyTorch：
选择 PyTorch 安装环境和版本

$ pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu124

Looking in indexes: https://download.pytorch.org/whl/nightly/cu124
Collecting torch
Downloading https://download.pytorch.org/whl/nightly/cu124/torch-2.4.0.dev20240605%2Bcu124-cp310-cp310-win_amd64.whl (2543.7 MB)
 ━━━━━━━━━━━━━━━╺━━━━━━━━━━━━━━━━━━━━━━━━ 1.0/2.5 GB 2.0 MB/s eta 0:13:16
...
Installing collected packages: tbb, mpmath, intel-openmp, sympy, networkx, mkl, torch, torchvision, torchaudio
Successfully installed intel-openmp-2021.4.0 mkl-2021.4.0 mpmath-1.2.1 networkx-3.2.1 sympy-1.12 tbb-2021.11.0 torch-2.4.0.dev20240605+cu124 torchaudio-2.2.0.dev20240605+cu124 torchvision-0.19.0.dev20240605+cu124