Closed beebopkim closed 5 months ago
Oh my bad. This is my fault. I happened to create a backup file called Makefile.local a while ago, but it was so old that I forgot about it, and it became a problem this time.
Ah alright. This was probably triggered because of https://github.com/LostRuins/koboldcpp/pull/896
Is it working fine for you now then?
Hi @beebopkim , could I trouble you to see if the latest concedo_experimental
builds and works on macOS with both regular Accelerate and also LLAMA_METAL=1
?
llama.cpp recently restructured openblas/accelerate out to a different backend, and also removed clblast in https://github.com/ggerganov/llama.cpp/pull/7735, so I've had to do a little bit of changes myself to preserve CLBlast. Just want to confirm that I didn't accidentally break macOS builds in the process.
I have no problems now. I renamed the backup in Makefile and the problem went away. There is no problem building with LLAMA_METAL=1 on macOS.
okay great, that means the Accelerate refactor works
Commit hash: 4fddbab0248cfda4ae3852e3448483de2f250c88
Build log:
Same error was happend with
LLAMA_METAL=1 make koboldcpp_default
too.