md/202510_RTX4090笔电操作记录.md

version: '3'
services:
  elasticsearch:
    image: swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/docker.elastic.co/elasticsearch/elasticsearch:7.17.28
    container_name: skw-es
    environment:
      - discovery.type=single-node
      - bootstrap.memory_lock=true
      - "ES_JAVA_OPTS=-Xms2g -Xmx2g"
    ulimits:
      memlock:
        soft: -1
        hard: -1
    ports:
      - "39876:9200"
    volumes:
      - /home/ss/vllm-py12/skw-es:/usr/share/elasticsearch/data

  oap:
    image: swr.cn-north-4.myhuaweicloud.com/ddn-k8s/skywalking.docker.scarf.sh/apache/skywalking-oap-server:9.5.0
    container_name: skw-oap
    depends_on:
      - elasticsearch
    links:
      - elasticsearch
    restart: always
    ports:
      - "38740:11800"
      - "34579:12800"
    environment:
      SW_STORAGE: elasticsearch
      SW_STORAGE_ES_CLUSTER_NODES: elasticsearch:39876
      SW_HEALTH_CHECKER: default
      JAVA_OPTS: "-Xms2g -Xmx2g"

  ui:
    image: swr.cn-north-4.myhuaweicloud.com/ddn-k8s/skywalking.docker.scarf.sh/apache/skywalking-ui:9.5.0
    container_name: skw-ui
    depends_on:
      - oap
    links:
      - oap
    restart: always
    ports:
      - "37658:8080"
    environment:
      SW_OAP_ADDRESS: http://oap:34579

# 因清华大学开源镜像站 HTTP/403 换了中科大的镜像站，配置信息存放在这里
cat /etc/apt/sources.list

# 安装 openssh 端口号是默认的 22 没有修改
sudo apt install openssh-server -y
sudo systemctl enable ssh
sudo systemctl start ssh

# 安装 NVDIA 显卡驱动和
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt-get update
sudo apt-get -y install cuda-toolkit-12-8
sudo apt-get install -y cuda-drivers
nvidia-smi

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

# 创建了一个新的目录，用于存储 vllm 使用的模型或其他文件
mkdir /home/ss/vllm-py12 && cd /home/ss/vllm-py12

# 用 conda 建了个新环境，以下 pip install 都是在该环境执行的
conda create -n vllm-py12 python=3.12 -y
conda activate vllm-py12

# 安装 vllm
pip install vllm -i http://mirrors.cloud.tencent.com/pypi/simple --extra-index-url https://download.pytorch.org/whl/cu128

# 安装 modelscope
pip install modelscope -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com

# 拉取 gpt-oss-20b 模型，由于显存不足，运行失败了
modelscope download --model openai-mirror/gpt-oss-20b --local_dir /home/ss/vllm-py12/gpt-oss-20b

# 下载了 Qwen3-0.6B 
modelscope download --model Qwen/Qwen3-0.6B --local_dir /home/ss/vllm-py12/qwen3-06b

# 运行 Qwen3-0.6B
nohup vllm serve /home/ss/vllm-py12/qwen3-06b \
    --host 0.0.0.0 \
    --port 8000 \
    --served-model-name Qwen3-0.6B \
    --tensor-parallel-size 1 \
    --dtype auto \
    --gpu-memory-utilization 0.9 \
    --max-model-len 32768 \
    --trust-remote-code \
	>> /home/ss/vllm-py12/vllm.log 2>&1 \
	& echo $! > /home/ss/vllm-py12/vllm.pid

# 安装了抓包工具 tshark 和 ngrep
sudo apt install ngrep
sudo apt-get install tshark

# 通过 java 脚本调用 tshark 提取关键日志
sudo nohup bash /home/ss/vllm-py12/tshark_bash.sh >> /home/ss/vllm-py12/tshark_bash.log 2>&1 & echo $! > /home/ss/vllm-py12/tshark_bash.pid

# 运行了1个定时任务脚本，清理 tshark 的临时文件并重启 java 脚本
sudo nohup /home/ss/vllm-py12/timer_bash.sh > /home/ss/vllm-py12/timer_bash.log 2>&1 & echo $! > /home/ss/vllm-py12/timer_bash.pid

# 杀死上面2个进程的命令
sudo kill -9 $(cat /home/ss/vllm-py12/timer_bash.pid)
sudo kill -9 $(cat /home/ss/vllm-py12/tshark_bash.pid)

# 清理日志
cd /home/ss/vllm-py12 && rm -rf timer_bash.log tshark_bash.log shark.log