Hi, Community,
I think it may be worth to discuss here any seen errors or improvements for the initial setup.
Please share your ideas here, too.
When my Spark arrived first point I had to find out was, that the expected “Hotspot Setup“ was not working at all. When I connected a monitor to see the reason it switched into the interactive setup. SLuckily I could just connect my HP docking station cable as “Keybord/Mouse“ and then I could finish this part.
Next step I spent quite a while to find out, that I had to install “NVIDIA Sync“ to connect to the Spark. That should really be mentioned on the quick start papers.
When I did the install of the Sync-Tool it throws an error that it could not access the file %userdir%\.ssh\config. SO I had to manually create that file first and then finally the setup worked.
After this I decided to install Open WebUI on my Spark. For this task I used the tool “MobaXterm” because it could combine SSH-Terminal with a file-browser in one package. Then I followed all instructions from the NVIDIA playbook. to install WebUI.
2 improvments for this part plus 1 warning:
a) It should be mentioned that you need to switch the Ollama-URL to http://127.0.0.1:11434 to work properly
b) for the custom script to install & connect to the OpenUI GUI you should better use this docker line to also be able to access the Ollama instanvce from other computers in the LAN:
docker run -d -p 12000:8080 -p 11434:11434 --gpus=all
-e OLLAMA_HOST=0.0.0.0
-v open-webui:/app/backend/data
-v open-webui-ollama:/root/.ollama
–name “${NAME}” --restart always “${IMAGE}” >/dev/null
also make sure to open the firewall port accordingly to make it work via this SSH-command:
sudo ufw allow 11434/tcp
c) the warning: dont try to be smart and change the open-webui:ollama to open-webui:latest. then you will not have the Ollama instance in the container and nothing will work ;-)
Anything else that is worth to share?
Please let the community (and NVIDIA) know…