【BUG】Error: llama runner process has terminated: exit status 127

本地私有化部署大模型，主流的工具是 Ollama。
使用以下指令部署：

curl -fsSL https://ollama.com/install.sh | sh

但是笔者本地报错了，查下gitbub 手动下载：

curl -L https://ollama.com/download/ollama-linux-amd64-rocm.tgz -o ollama-linux-amd64-rocm.tgz
sudo tar -C /usr -xzf ollama-linux-amd64-rocm.tgz

或者使用[三方镜像](https://docker.aityp.com/image/docker.io/ollama/ollama:rocm） rocm 是支持GPU的意思。

docker pull swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/ollama/ollama:rocm
docker tag  swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/ollama/ollama:rocm  docker.io/ollama/ollama:rocm

安装完毕执行：

sh-4.2# ollama run glm4
Error: llama runner process has terminated: exit status 127

由于日志比较少，打开docker的日志分析：

[GIN] 2024/11/15 - 07:41:49 | 200 |     283.749µs |       127.0.0.1 | HEAD     "/"
[GIN] 2024/11/15 - 07:41:49 | 200 |   17.617406ms |       127.0.0.1 | POST     "/api/show"
time=2024-11-15T07:41:49.430Z level=INFO source=sched.go:715 msg="new model will fit in available VRAM in single GPU, loading" model=/root/.ollama/models/blobs/sha256-b506a070d1152798d435ec4e7687336567ae653b3106f73b7b4ac7be1cbc4449 gpu=GPU-c0da87a9-e0be-be71-6ee5-496aa7f0d6d0 parallel=4 available=9513730048 required="6.2 GiB"
time=2024-11-15T07:41:49.431Z level=INFO source=memory.go:309 msg="offload to cuda" layers.requested=-1 layers.model=41 layers.offload=41 layers.split="" memory.available="[8.9 GiB]" memory.required.full="6.2 GiB" memory.required.partial="6.2 GiB" memory.required.kv="320.0 MiB" memory.required.allocations="[6.2 GiB]" memory.weights.total="4.6 GiB" memory.weights.repeating="4.1 GiB" memory.weights.nonrepeating="485.6 MiB" memory.graph.full="561.0 MiB" memory.graph.partial="789.6 MiB"
time=2024-11-15T07:41:49.445Z level=INFO source=server.go:391 msg="starting llama server" cmd="/tmp/ollama3667284545/runners/cuda_v12/ollama_llama_server --model /root/.ollama/models/blobs/sha256-b506a070d1152798d435ec4e7687336567ae653b3106f73b7b4ac7be1cbc4449 --ctx-size 8192 --batch-size 512 --embedding --log-disable --n-gpu-layers 41 --parallel 4 --port 33511"
time=2024-11-15T07:41:49.445Z level=INFO source=sched.go:450 msg="loaded runners" count=1
time=2024-11-15T07:41:49.445Z level=INFO source=server.go:591 msg="waiting for llama runner to start responding"
time=2024-11-15T07:41:49.446Z level=INFO source=server.go:625 msg="waiting for server to become available" status="llm server error"
/tmp/ollama3667284545/runners/cuda_v12/ollama_llama_server: error while loading shared libraries: libcudart.so.12: cannot open shared object file: No such file or directory
time=2024-11-15T07:41:49.698Z level=ERROR source=sched.go:456 msg="error loading llama server" error="llama runner process has terminated: exit status 127"
[GIN] 2024/11/15 - 07:41:49 | 500 |  654.604793ms |       127.0.0.1 | POST     "/api/chat"
time=2024-11-15T07:41:54.826Z level=WARN source=sched.go:647 msg="gpu VRAM usage didn't recover within timeout" seconds=5.128305852 model=/root/.ollama/models/blobs/sha256-b506a070d1152798d435ec4e7687336567ae653b3106f73b7b4ac7be1cbc4449
time=2024-11-15T07:41:55.077Z level=WARN source=sched.go:647 msg="gpu VRAM usage didn't recover within timeout" seconds=5.378774395 model=/root/.ollama/models/blobs/sha256-b506a070d1152798d435ec4e7687336567ae653b3106f73b7b4ac7be1cbc4449
time=2024-11-15T07:41:55.326Z level=WARN source=sched.go:647 msg="gpu VRAM usage didn't recover within timeout" seconds=5.628515531 model=/root/.ollama/models/blobs/sha256-b506a070d1152798d435ec4e7687336567ae653b3106f73b7b4ac7be1cbc4449

关键问题日志：
/tmp/ollama3667284545/runners/cuda_v12/ollama_llama_server: error while loading shared libraries: libcudart.so.12: cannot open shared object file: No such file or directory

初步分析是缺少cuda的驱动，版本是V12。到英伟达官网（需要注册）查到对应的驱动，因为rocm 底层操作系统镜像是centos，所有高版本的工具都没有了。下载地址

执行以下指令：

wget https://developer.download.nvidia.com/compute/cuda/12.0.0/local_installers/cuda_12.0.0_525.60.13_linux.run
sudo sh cuda_12.0.0_525.60.13_linux.run

出现的终端，输入accpet，然后下一个界面出现的“X”按Enter键取消，选择“Install”进行安装。
然后设置环境变量：

vim  ~/.bashrc

export PATH=$PATH:/usr/local/cuda-12.0/bin/

source  ~/.bashrc

验证

[root@039e8aa5e3fc /]# nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Mon_Oct_24_19:12:58_PDT_2022
Cuda compilation tools, release 12.0, V12.0.76
Build cuda_12.0.r12.0/compiler.31968024_0

执行完毕再次运行模型:

[root@039e8aa5e3fc /]# ollama run glm4
>>> hello
Hello ! How can I assist you today?

附件

Centos 替换系统镜像

在CentOS中，如果你想替换系统镜像源，通常是为了快速下载或更新系统软件包。以下是一个简单的步骤指南和示例代码，用于替换默认的CentOS镜像源为阿里云镜像源。
1.备份当前的yum仓库配置。

sudo cp -a /etc/yum.repos.d /etc/yum.repos.d.backup

2.清除原有的yum仓库配置。

sudo rm -f /etc/yum.repos.d/*.repo

3.下载阿里云的CentOS镜像源配置文件。

sudo curl -o /etc/yum.repos.d/CentOS-Base.repo http://mirrors.aliyun.com/repo/Centos-7.repo

4.除yum缓存并生成新的缓存。

sudo yum clean allsudo yum makecache

5.更新系统（可选）。

sudo yum update

以上步骤将会将你的CentOS系统的yum镜像源更换为阿里云的镜像源，并生成新的缓存文件，以便于快速下载和安装软件包。如果你使用的是CentOS 8或其他版本，请确保下载对应版本的镜像源配置文件。