【信创】外网aarch64终端安装deepseek Ragflow 迁移至内网银河麒麟服务器

真实环境实验外网飞腾D2000 4核16G / UOS107064K页内网FT2000 128G / 麒麟V10矛系统完全断网【信创】外网aarch64终端安装deepseek Ragflow 迁移至内网银河麒麟服务器第一阶段外网 D2000 UOS 1070 操作联网步骤1手动进入ROOT必须先做打开终端输入sudosu现在你是 # 最高权限。步骤2创建第一个脚本 01env_build.sh新建文件nano01env_build.sh粘贴下面完整修复好的代码无任何sudo错误#!/bin/bashset-eaptupdateaptinstall-ygitwgetcurlca-certificates python3-pip# Docker静态包中科大可用wgethttps://mirrors.ustc.edu.cn/docker-ce/linux/static/stable/aarch64/docker-26.0.2.tgztar-zxfdocker-26.0.2.tgzcpdocker/* /usr/local/bin# 写入docker systemd服务cat/etc/systemd/system/docker.serviceSERVICE [Unit] DescriptionDocker Application Container Engine Afternetwork-online.target [Service] Typenotify ExecStart/usr/local/bin/dockerd ExecReload/bin/kill -s HUP $MAINPID TimeoutSec0 RestartSec2 Restartalways [Install] WantedBymulti-user.target SERVICEsystemctl daemon-reload systemctl startdockersystemctlenabledocker# 先停止正在运行的dockersystemctl stopdockersystemctl stop containerd# 确认docker进程全部退出pkilldockerdpkillcontainerd# 处理composepip3installdocker-composeln-sf/usr/local/bin/docker-compose /usr/bin/docker-compose# 重新启动docker服务并验证systemctl daemon-reload systemctl startdockerdocker-vdocker-compose-vecho【OK】Docker 26.0.2 安装成功保存CtrlO 回车退出CtrlX小提示以后不要完整重跑整个脚本Docker程序正在运行时不能覆盖。现在只要两条命令输出版本号01这一步就彻底完成直接进入构建镜像的阶段。步骤3赋予权限并运行chmodx 01env_build.sh ./01env_build.sh✅ Docker 安装完成步骤4创建 02build_ragflow.sh构建镜像nanobuild_ragflow.sh粘贴#!/bin/bashset-ecd/root# 拉取源码gitclone https://github.com/infiniflow/ragflow.gitcdragflow# 构建ARM64镜像dockerbuild-tragflow:arm64.# 拉取中间件dockerpull postgres:15-alpinedockerpull redis:7-alpinedockerpull nginx:alpineecho【OK】RAGFlow镜像构建完成运行chmodx build_ragflow.sh ./build_ragflow.sh步骤5创建 install_ollama_build.sh安装Ollama下载模型nanoinstall_ollama_build.sh粘贴#!/bin/bashset-e# 安装ollamawgethttps://ollama.com/download/ollama-linux-arm64mvollama-linux-arm64 /usr/local/bin/ollamachmodx /usr/local/bin/ollama# 写入系统服务cat/etc/systemd/system/ollama.serviceEOF [Unit] DescriptionOllama Service Afternetwork.target [Service] EnvironmentOLLAMA_HOST0.0.0.0 ExecStart/usr/local/bin/ollama serve Restartalways [Install] WantedBymulti-user.target EOFsystemctl daemon-reload systemctl start ollama systemctlenableollama# 安装下载工具pip3installhuggingface_hub[cli]exportHF_ENDPOINThttps://hf-mirror.com# 新建模型目录mkdir-p/root/modelscd/root/models# 下载14B模型huggingface-cli download bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF\--include*Q4_K_M.gguf--local-dir ./# 下载向量模型huggingface-cli download bartowski/bge-small-zh-v1.5-GGUF\--include*Q4_K_M.gguf--local-dir ./# 导入大模型catModelfile-14bMF FROM ./DeepSeek-R1-Distill-Qwen-14B.Q4_K_M.gguf PARAMETER temperature 0.6 PARAMETER num_ctx 8192 MFollama create deepseek-r1-14b-fModelfile-14b# 导入向量模型catModelfile-bgeMF2 FROM ./bge-small-zh-v1.5.Q4_K_M.gguf MF2ollama create bge-small-zh-v1.5-fModelfile-bge ollama listecho【OK】模型导入完成禁止对话测试16G内存会崩运行chmodx install_ollama_build.sh ./install_ollama_build.sh步骤6创建 export_offline.sh打包全部离线包nanoexport_offline.sh粘贴#!/bin/bashset-emkdir-p/root/offline_bundle# 导出全部docker镜像dockersave-o/root/offline_bundle/ragflow_bundle.tar ragflow:arm64 postgres:15-alpine redis:7-alpine nginx:alpine# 打包部署文件mkdir-p/root/deploy_packcp/usr/local/bin/ollama /root/deploy_pack/cp-r/root/.ollama /root/deploy_pack/ollama_datacp/root/ragflow/docker/.env /root/deploy_pack/cp/root/ragflow/docker/docker-compose.yml /root/deploy_pack/tar-zcvf/root/offline_bundle/deploy_pack.tar.gz-C/root/deploy_pack.cp/root/docker-26.0.2.tgz /root/offline_bundle/# MD5校验cd/root/offline_bundle md5sum ragflow_bundle.tar deploy_pack.tar.gz docker-26.0.2.tgzfile_md5.txtecho【OK】全部离线包打包完成路径/root/offline_bundlels-lh/root/offline_bundle运行chmodx export_offline.sh ./export_offline.sh✅【外网所有工作到此结束】你得到3个超级重要文件ragflow_bundle.tar 全部Docker镜像deploy_pack.tar.gz Ollama完整模型配置docker-26.0.2.tgz 内网安装Docker用重点外网绝对不做问答测试16G必崩第二阶段U盘拷贝U盘必须exFAT把/root/offline_bundle全部拷走第三阶段内网 FT2000 128G 麒麟断网部署最终运行1. 内网终端先切rootsudosu2. 挂载U盘mkdir-p/mnt/usb lsblkmount/dev/sdb1 /mnt/usb3. 内网安装Docker脚本 env_prod.shnanoenv_prod.sh粘贴#!/bin/bashset-emkdir-p/datacp-r/mnt/usb/offline_bundle /data/cd/data/offline_bundletar-zxfdocker-26.0.2.tgzcpdocker/* /usr/bincat/etc/systemd/system/docker.serviceSERVICE [Unit] DescriptionDocker Application Container Engine Afternetwork-online.target [Service] Typenotify ExecStart/usr/bin/dockerd ExecReload/bin/kill -s HUP $MAINPID TimeoutSec0 RestartSec2 Restartalways [Install] WantedBymulti-user.target SERVICEsystemctl daemon-reload systemctl startdockersystemctlenabledockerdocker-vecho【OK】内网Docker安装完成运行chmodx env_prod.sh ./env_prod.sh4. 内网部署全套业务脚本 deploy_prod.shnanodeploy_prod.sh粘贴#!/bin/bashset-ecd/data/offline_bundle md5sum-cfile_md5.txt# 导入镜像dockerload-iragflow_bundle.tar# 解压部署包cd/datatar-zxvf/data/offline_bundle/deploy_pack.tar.gz# 恢复ollamacp/data/ollama /usr/local/bin/ollamachmodx /usr/local/bin/ollamacp-r/data/ollama_data /root/.ollama# 写入服务cat/etc/systemd/system/ollama.serviceEOF [Unit] DescriptionOllama Service Afternetwork.target [Service] EnvironmentOLLAMA_HOST0.0.0.0 ExecStart/usr/local/bin/ollama serve Restartalways [Install] WantedBymulti-user.target EOFsystemctl daemon-reload systemctlenableollama systemctl start ollama ollama listecho【OK】内网模型环境全部恢复完成无需重新创建模型运行chmodx deploy_prod.sh ./deploy_prod.sh5. 修改内网.env配置vim/data/.env改内网IP、数据库参数6. 启动项目cd/datadocker-composeup-d最终成功浏览器打开内网IP全套RAGFlow14B大模型离线运行总结核心真相脚本里不能写 sudo su / sudo -i——会卡死人所有脚本执行前手动sudo su切root——永远不卡外网D2000只打包、不测试——防止OOM内网FT2000128G专门跑正式业务——稳如服务器64K→64K页大小零兼容错误比从阿里云g8r用虚拟机更稳