chore: update flags to stay consistent with llama.cpp

Author: jahfer

README.md

@@ -28,8 +28,8 @@ There are several installation options:
 # use OpenBLAS
 $ gem install llama_cpp -- --with-openblas
 
-# use cuBLAS
-$ gem install llama_cpp -- --with-cublas
+# use CUDA
+$ gem install llama_cpp -- --with-cuda
 ```
 
 Those options are defined in [extconf.rb](https://github.com/yoshoku/llama_cpp.rb/blob/main/ext/llama_cpp/extconf.rb) by with_config method.

ext/llama_cpp/extconf.rb

@@ -15,7 +15,8 @@
 make_envs << ' LLAMA_NO_ACCELERATE=1' if with_config('no-accelerate')
 make_envs << ' LLAMA_OPENBLAS=1' if with_config('openblas')
 make_envs << ' LLAMA_BLIS=1' if with_config('blis')
-make_envs << ' LLAMA_CUBLAS=1' if with_config('cublas')
+make_envs << ' LLAMA_CUBLAS=1' if with_config('cublas') # Deprecated, use --with-cuda instead
+make_envs << ' LLAMA_CUDA=1' if with_config('cuda')
 make_envs << ' LLAMA_CLBLAST=1' if with_config('clblast')
 make_envs << ' LLAMA_HIPBLAS=1' if with_config('hipblas')
 make_envs << ' LLAMA_MPI=1' if with_config('mpi')