bug: Vulkan Backend not detected after update

[engrtipusultan]

I have updated to Jan 0.6.8 now it has stopped showing my igpu

Whereas LMstudio can see it properly.

Is there a way to fix it ?

[qnixsynapse]

Yeah, integrated GPU was disabled in this update because the vulkan backend was found to be not very stable on iGPUs.

In the next update, we will at-least show options in the llamacpp backend settings so that users can switch to it. But crashes and other problems can happen.

[engrtipusultan]

I dont agree that it is unstable lmstudio uses it. Also I have been using llama.cpp prebuilt binary for ubuntu. It only crashes if you try to run model bigger then your memory. Lmstudio has options to put guard rails. On lmstudio it does not even crash

[LazyYuuki]

Hey @engrtipusultan, I have added your suggestion above as one of the thing to work toward for 0.7.0. Do look out for it soon!

[k9janer]

For me the error show a bit different:

This is with an AMD Ryzen 9 8945 HS /w Radeon 780M Graphics using OpenSuse Tumbleweed Slowroll.

jan also doesn't work at all for me using a local model. It seems to be stuck trying to load a suitable llama.ccp and the spinner is constantly loading. Is this related to this IGPU detection or am I running into a different issue and I should open a new issue?

Personally I've also never had any issues with the IGPU using LM Studio. To load models bigger than half my RAM size into IGPU "VRAM" I had to increase GTT size with a kernel parameter, but now I can easily and reliably load models up to 24GB with my 32GB RAM Laptop and I've had zero problems so far.

This is the hardware tab in LM Studio (ignore the incorrect 16GB VRAM, it can use up to 24GB VRAM):

[LazyYuuki]

hi @k9janer I think this is a different problem, your backend is not even downloaded at all. There is probable some problem with your connection to Github to download the backend from llama.cpp

Do try to restart Jan and help us take a screenshot of the page on Model Provider -> Llama.cpp -> Backend

[k9janer]

I've restarted jan a few times now but llama.cpp always shows as loading. I don't have any issues accessing github in my browser and the update check in jan also works without any issues.

The app log: k9janer jan.log

[LazyYuuki]

get jan extensions, path: "/home/k9janer/.local/share/Jan/data/extensions/extensions.json"
[2:40:00 PM]
DEBUG
checking for updates https://github.com/menloresearch/jan/releases/latest/download/latest.json
[2:40:00 PM]
DEBUG
starting new connection: https://github.com/
[2:40:00 PM]
DEBUG
starting new connection: https://release-assets.githubusercontent.com/
[2:40:00 PM]
ERROR
Unable to initialize NVML: a libloading error occurred: libnvidia-ml.so.1: cannot open shared object file: No such file or directory
[2:40:00 PM]
ERROR
Failed to get NVIDIA GPUs: an internal driver error occured
[2:40:00 PM]

@qnixsynapse can you help me check @k9janer log? I saw this as the main culprit but I am not so sure what might have caused it