I tried two Llamafiles (llava-v1.5-7b-q4.llamafile and mistral-7b-instruct-v0.2.Q4_0.llamafile) both about 4 GB in size.
(run fine on ryzen 9 and
...