I erroneously filed the bug against llama.cpp, instead of ggml.
I'm lowering the severity because this test passes on ci.debian.net's
arm64. This could actually be a hardware-specific issue with my device
(Orion O6).
Furthermore, only a single test is affected:
MUL_MAT] ERR = 0.193812193 > 0.000500000
MUL_MAT(type_a=f32,type_b=f32,m=64,n=77,k=77,bs=[12,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1):
FAIL
I initially thought of skipping this one test, but it's better to keep
it for now, so that I can re-run it on other hardware.
Best,
Christian