How to automatically start lm-studio in linux?

Posted by badabimbadabum2@reddit | LocalLLaMA | View on Reddit | 12 comments

I have tried to start the lms studio with a bash script in crontab. But nothing seems to work. Even with ssh connection to the server, lms server start wont start the server. lms server start Starting server... Waking up LM Studio service... E Failed to start the server on port 1234 But when goin locally the the server, that same command works from the command line: lms server start. What is this? some kind of protection not to automatically start it?

Reply to Post

12 Comments

[-]

huzbum@reddit

I was also very frustrated that I couldn't use LM studio \*actually\* headless without opening LM Studio in the GUI first. I often remote into my machine, and I rebooted without thinking... argh now I can't run lms. I found this guide to setting up a service so lms actually works: [https://run.tournament.org.il/running-headless-lm-studio-on-ubuntu/](https://run.tournament.org.il/running-headless-lm-studio-on-ubuntu/)

[-]

Relevant-Guidance-95@reddit

Hi, I am having a similar problem. The server does not start when that command is executed from a ssh connection, and it starts fine when the connection includes a X pipe (ssh -Y). So it looks like the problem is that that command still needs the X graphical interface for something, even if it does not open any window. It is probably a bug as there is no reason why that process needs X. So it looks like in your case that process in the crontab does not have access to the X system, and that is the reason it fails.

[-]

badabimbadabum2@reddit (OP)

Yes, I gave up with lm-studio long time ago, its not made for server workloads anyway.

[-]

skidmarksteak@reddit

If it's working from the command line then it's most likely related to environment variables. Start it in the background like this: screen -dmS lm-studio /bin/bash -c '/users/badalimb/something/lms server start' You might need to install the "screen" command, which you can do by typing "sudo apt-get install screen" or "dnf install screen" depending on your distro. If it's working (it should) you can bring it back to the foreground by typing screen -r lm-studio. As for autostarting it ... type 'cat /etc/os-release' so I can help you with setting it up as a service and autostart it.

[-]

badabimbadabum2@reddit (OP)

PRETTY_NAME="Ubuntu 24.04.1 LTS" NAME="Ubuntu" VERSION_ID="24.04" VERSION="24.04.1 LTS (Noble Numbat)" VERSION_CODENAME=noble ID=ubuntu ID_LIKE=debian HOME_URL="https://www.ubuntu.com/" SUPPORT_URL="https://help.ubuntu.com/" BUG_REPORT_URL="https://bugs.launchpad.net/ubuntu/" PRIVACY_POLICY_URL="https://www.ubuntu.com/legal/terms-and-policies/privacy-policy" UBUNTU_CODENAME=noble LOGO=ubuntu-logo lms server start --verbose D Reading last status from /home/ubuntu/.cache/lm-studio/.internal/http-server-config.json D Read from last status: port=1234 D Attempting to start the server on port 1234... I Starting server... D Resolved serverCtlPath: /home/ubuntu/.cache/lm-studio/.internal/http-server-ctl.json D Making sure directory exists: /home/ubuntu/.cache/lm-studio/.internal D Writing control object to /home/ubuntu/.cache/lm-studio/.internal/http-server-ctl.json: { type: 'start', port: 1234, cors: false } D Attempt 1: File has not been cleared D Attempt 2: File has not been cleared D Attempt 3: File has not been cleared D Attempt 4: File has not been cleared D Attempt 5: File has not been cleared D Attempt 6: File has not been cleared D Attempt 7: File has not been cleared D Attempt 8: File has not been cleared D Attempt 9: File has not been cleared D Attempt 10: File has not been cleared I Waking up LM Studio service... D Resolved appInstallLocationPath: /home/ubuntu/.cache/lm-studio/.internal/app-install-location.json D Read executable pointer: { path: '/usr/local/squashfs-root/lm-studio', argv: [ '/usr/local/squashfs-root/lm-studio', '--minimized', '--run-as-service' ], cwd: '/usr/local/squashfs-root' } D Spawning process: { path: '/usr/local/squashfs-root/lm-studio', args: [ '--minimized', '--run-as-service' ], cwd: '/usr/local/squashfs-root' } D Process spawned D LM Studio service is running. D Attempt 1: File has not been cleared D Attempt 2: File has not been cleared D Attempt 3: File has not been cleared D Attempt 4: File has not been cleared D Attempt 5: File has not been cleared D Attempt 6: File has not been cleared D Attempt 7: File has not been cleared D Attempt 8: File has not been cleared D Attempt 9: File has not been cleared D Attempt 10: File has not been cleared E Failed to start the server on port 1234

[-]

skidmarksteak@reddit

In this case login to the terminal where it works correctly and start it in a screen session, let's see if it works. Just type this: screen lms server start If this works correctly then press ctrl+d to "detach" the session, it will keep running in the background. You can bring it back to the foreground by typing screen -r. Let me know if this works - it really should - so we can debug further. Debian or Ubuntu makes no difference in this regard, you can rule it out.

[-]

ali0une@reddit

You need version 0.3.5 or above and have to launch it with /path/to/LM\_Studio.AppImage --run-as-service Then you'll have its icon in the notification area and you'll be able to use lms.

[-]

badabimbadabum2@reddit (OP)

But how can I automatically after reboot start lm-studio? So when I run in the command line: path/to/lm-studio --run-as-service the lms status shows Server: On. Thats what I want to automate. But when I create a .sh script and try to put it in the crontab with the reboot command @reboot /scripts/start-lms.sh It never starts after reboot, the **lms status** \-command shows for the server: OFF. Also tried to put in the crontab path/to/lm-studio --run-as-service but nothing works.

[-]

EmilPi@reddit

I am not sure that this is related, but until recent version, it was necessary that LM Studio runs in graphical environment, only then you can start server in terminal. In recent version, they claimed "headless mode", but I didn't check if it works, as I since a few months I migrated to llama.cpp + systemd service completely. It is not that difficult.

[-]

EmilPi@reddit

https://preview.redd.it/s73fkynrxqzd1.png?width=683&format=png&auto=webp&s=e8395d819366d75172b557019656a7dc6f0dccce Unfortunately, I cannot copypaste code because reddit thinks it is some executable malicious cross-forgery scripting attack, even it is just systemd service code.

[-]

skidmarksteak@reddit

You can copypaste code on reddit using markdown like this: \`\`\` <code goes here> \`\`\`

[-]

EmilPi@reddit

What you need for this, is first to install llama.cpp . I did it from source, by cloning llama.cpp, entering its folder and running (no need to mention I have CUDA installed already) cmake -B build -DGGML_CUDA=ON -DGGML_CUDA_F16=ON cmake --build build --config Release --parallel 16