bug: BUG on newest jan v5.0, it set the default to gpu instead of cpu

[abaynp]

Describe the bug
Before I use Jan v4.x and also update to v5.0, with a cortex model, and it works flawlessly
But when I clean install v5.0, it cannot start the model. I manage to get it works by switching to experimental mode, then turn on the vulkan (which obviously I don't have any amd gpu), run into the same problem, then switch off the vulkan and the experimental mode, then the cortex model will start as normal.

I look at the logs, I suprised that it try to load something from win-cuda folder. Look at it closer turn out it sets to run on gpu mode ("run_mode":"gpu") I don't know how the way the app recognize my system, but my system is just another intel based thinkpad without any other gpu installed.

this is a part of the log
[APP]::{"notify":true,"run_mode":"gpu","nvidia_driver":{"exist":false,"version":""},"cuda":{"exist":true,"version":"11"},"gpus":[],"gpu_highest_vram":"","gpus_in_use":[""],"is_initial":false,"vulkan":false}

I don't know why it sets to gpu on default installation, may be because it found a cuda which I'm not really sure, is Intel® HD Graphics 620 support cuda? I believe it is not.

But when I comeback from vulkan to non vulkan, it successfully go to cpu mode, rather than gpu mode from clean install or factory reset.

I know it can be fix by manually edit a config.json or something, but this still in fact a BUG

app.log

[Van-QA]

We are in the process of restructuring Jan to use Cortex, where the hardware information will be acquired from the library side, and we will follow up on the problem once https://github.com/janhq/cortex-internal/issues/20 is resolved, thank you

[freelerobot]

What's the source of the issue @louis-jan ? Is it related to janhq/cortex.cpp#1142

[louis-menlo]

@0xSage This was a bug related to GPU detection, which could be triggered by the presence of nvidia-smi installed on the user's machine or weird response. The upcoming hardware detection feature in cortex-cpp might fix this.

[dan-menlo]

This issue is part of Cortex's Hardware Detection API, and is queued for Sprint 23. We will track this issue as part of that epic.

janhq/cortex.cpp#1165