-
It took me an hour, but I couldn't get it to work. Need help. Also, if openBLAS improves performance, even if not much, why it isn't used by default? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 10 replies
-
I have done this, I'll try to explain. First, get w64devkit Then, get OpenBLAS Put w64devkit somewhere you like, no need to set up anything else like PATH, there is just one executable that opens a shell, from there you can build llama.cpp with Now take the OpenBLAS release and from there copy
The files are going to be big because of static linking, you can also copy You can also compile OpenBLAS yourself, for that I recommend the Fortran version of w64devkit. |
Beta Was this translation helpful? Give feedback.
I have done this, I'll try to explain.
First, get w64devkit
w64devkit-1.18.0.zip
Then, get OpenBLAS
OpenBLAS-0.3.23-x64.zip
Put w64devkit somewhere you like, no need to set up anything else like PATH, there is just one executable that opens a shell, from there you can build llama.cpp with
make
as usual.Now take the OpenBLAS release and from there copy
lib/libopenblas.a
intow64devkit/x86_64-w64-mingw32/lib
and frominclude
copy all the .h files tow64devkit/x86_64-w64-mingw32/include
make LLAMA_OPENBLAS=1
should work now.The files are going to be big because of static linking, you can also copy
libopenblas.dll.a
but then you have to keep the .dll file around where OS can load it.You ca…