Re: Building llama.cpp for AMD GPU using only Debian packages?
[Petter Reinholdtsen]
> Sadly this build crashed with "1307996 FAILED TESTS". Only two tests
> more succeeded with no f16c. :(
On the more positive side, building with DEB_BUILD_OPTIONS=nocheck to
disable the self testing (which still took 8 hours), I got llama.cpp
working with the 4.8GiB sized dolphin-2.2.1-mistral-7b.Q5_K_M.gguf
model, providing 75 tokes/second. So it look like a one line change
fixed the problem for my old CPU. :)
Perhaps the patch to disable f16c should be included in unstable as
well?
--
Happy hacking
Petter Reinholdtsen
Reply to: