Gpu support #1267

I have low end laptop with i5-8250u and Nvidia mx 150 with 4 gb vram (equivalent to gt1030) and 8gb ram
I used whisper.cpp to transcribe a 43 minutes video and it takes about 30 minutes, then I tried Const-me/Whisper which supports GPU inference of whisper.cpp and it only takes 3 minutes (that is 10x faster)
Can we use the same method for llama? cpp especially with the similarity of whisper.cpp and llama.cpp? I am not good at these things but can someone look at the repo and start the discussion of how easy will it be to implement the same method?