DGX Spark 安裝 Text to Knowledge Graph

安裝 Text to Knowledge Graph

參考 https://build.nvidia.com/spark/txt2kg/instructions

$ git clone https://github.com/NVIDIA/dgx-spark-playbooks

$ cd dgx-spark-playbook/nvidia/txt2kg/assets

$ ./start.sh

瀏覽器 http://localhost:3001

$ ./stop.sh

因為同時使用了 Open WebUI with Ollama docker, 切換 compose_ollama_data 到 open-webui-ollama

$ docker volume ls

$ docker volume inspect compose_ollama_data

[

    {

        "CreatedAt": "2025-12-19T14:38:37+08:00",

        "Driver": "local",

        "Labels": {

            "com.docker.compose.config-hash": "b60a4e44fe9b008057f3eaff8c4477427e0db99c0c9a70285f81b92ba016830d",

            "com.docker.compose.project": "compose",

            "com.docker.compose.version": "2.40.0",

            "com.docker.compose.volume": "ollama_data"

        },

        "Mountpoint": "/var/lib/docker/volumes/compose_ollama_data/_data",

        "Name": "compose_ollama_data",

        "Options": null,

        "Scope": "local"

    }

]

$ cd dgx-spark-playbooks/nvidia/txt2kg/assets/deploy/compose

$ cp docker-compose.yml docker-compose.yml.bak

$ vi docker-compose.yml

將

  ollama:

    volumes:

      - ollama_data:/root/.ollama

改成

  ollama:

    volumes:

      - open-webui-ollama:/root/.ollama

將

volumes:

  ollama_data:

改成

volumes:

  open-webui-ollama:

    external: true

查詢 ollama-compose docker 目前使用的 volume

$ docker inspect ollama-compose --format '{{ json .Mounts }}'

把 volume 掛進暫時容器，並清空內容

$ docker run --rm \

  -v compose_ollama_data:/data \

  alpine \

  sh -c "rm -rf /data/*"

確保沒有容器在用

$ docker ps -a --filter volume=compose_ollama_data

刪除 volume

docker volume rm compose_ollama_data

$ export OLLAMA_MODEL=gpt-oss:20b

$ docker exec ollama-compose ollama list

$ docker exec ollama-compose ollama pull gpt-oss:20b