CUDA doesnt work when compiling for windows, Unity 6, Nvidia

[captainzero93]

Describe the bug

Correctly loads the model when I am in the Unity play mode but when compiled it reverts to CPU and is too slow to load. I've tried to clean the projects, re-import LLMUnity, I've tried taking the .dlls from the streamed folders with the build and placing them with the compiled .exe but nothing seems to work !

May be worth noting I've enabled 'extras' since it's an iquant model (LLama3 Based )

No errors in the build log or warnings!

LLMUnity version

v2.4.1

Operating System

Windows

Specs:

Windows 11 LTSC
Unity 6
4070 12gb VRAM ( GGUF model is only 2gb)
32GB DDR4

note; I also tried installing the specific version of CUDA locally but I don't think this is the issue ( and it did not resolve the issue with the windows build )

[amakropoulos]

Yes you are right, I realised that last week.
The issue was actually very simple:
In the build I deploy only the "full" DLLs but I don't save that it should use those instead of the default DLLs.
It is fixed with this PR.
It will be released with v2.4.2 this week.