最近再部署大模型玩

1.问题出现

docker版本如下

创建文件docker-compose.yml文件如下

name: 'ollama'
services:
  ollama:
    #restart: always
    image: ollama/ollama
    container_name: ollama13
    runtime: nvidia
    environment:
      - TZ=Asia/Shanghai
      - NVIDIA_VISIBLE_DEVICES=all
    networks:
      - ai-tier
    ports:
      - "11745:11434"
    volumes:
      - ./data:/root/.ollama
networks:
  ai-tier:
    name: ai-tier
    driver: bridge

启动容器 

docker compose up -d

报错:no compatible GPUs were discovered

no nvidia devices detected by library /usr/lib/x86 64-linux-gnu/libcuda.so.550.135

2.测试宿主机

curl -fsSL https://ollama.com/install.sh | sh
# 下载完成之后执行
ollama run llama3.2

在宿主机启动容器测试,可以正常调动GPU,说明驱动没有问题

3. 排查问题

测试能否调用GPU

docker run --rm --gpus all  nvidia/cuda:12.0.1-runtime-ubuntu22.04 nvidia-smi

[root@worker1 ~]# docker run -it --rm --gpus all nvidia/cuda:12.4.0-base-ubuntu22.04 nvidia-smi
Unable to find image 'nvidia/cuda:12.4.0-base-ubuntu22.04' locally
12.4.0-base-ubuntu22.04: Pulling from nvidia/cuda
bccd10f490ab: Pull complete 
edd1dba56169: Pull complete 
e06eb1b5c4cc: Pull complete 
7f308a765276: Pull complete 
3af11d09e9cd: Pull complete 
Digest: sha256:80d4d9ac041242f6ae5d05f9be262b3374e0e0b8bb5a49c6c3e94e192cde4a44
Status: Downloaded newer image for nvidia/cuda:12.4.0-base-ubuntu22.04
Failed to initialize NVML: Unknown Error

根据报错

修改配置文件

vim /etc/nvidia-container-runtime/config.toml

将图中no-cgroups=true改成

no-cgroups=false

此参数对任务使用的资源（内存，CPU，磁盘等资源）总额进行限制

 修改完成后,重启docker服务

systemctl restart docker

4.验证是否解决

docker run --rm --gpus all  nvidia/cuda:12.0.1-runtime-ubuntu22.04 nvidia-smi

至此问题解决