diff --git a/Makefile b/Makefile
index dea93d83e..cf110833c 100644
--- a/Makefile
+++ b/Makefile
@@ -1,4 +1,4 @@
-default: koboldcpp koboldcpp_noavx2 koboldcpp_openblas koboldcpp_openblas_noavx2 koboldcpp_clblast 
+default: koboldcpp koboldcpp_noavx2 koboldcpp_openblas koboldcpp_openblas_noavx2 koboldcpp_clblast koboldcpp_cublas
 simple: koboldcpp koboldcpp_noavx2
 dev: koboldcpp_openblas
 
@@ -45,8 +45,8 @@ endif
 #
 
 # keep standard at C11 and C++11
-CFLAGS   = -I.              -Ofast -DNDEBUG -std=c11   -fPIC
-CXXFLAGS = -I. -I./examples -Ofast -DNDEBUG -std=c++11 -fPIC
+CFLAGS   = -I.              -I./include -I./include/CL -Ofast -DNDEBUG -std=c11   -fPIC
+CXXFLAGS = -I. -I./examples -I./include -I./include/CL -Ofast -DNDEBUG -std=c++11 -fPIC
 LDFLAGS  =
 
 # these are used on windows, to build some libraries with extra old device compatibility
@@ -55,7 +55,7 @@ BONUSCFLAGS2 =
 
 OPENBLAS_FLAGS = -DGGML_USE_OPENBLAS -I/usr/local/include/openblas
 CLBLAST_FLAGS = -DGGML_USE_CLBLAST -DGGML_USE_OPENBLAS -I/usr/local/include/openblas
-CUBLAS_FLAGS = -DGGML_USE_CUBLAS -I/usr/local/cuda/include
+CUBLAS_FLAGS = -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I./include/cuda -I./include/cuda/crt
 
 #lets try enabling everything
 CFLAGS   += -pthread -s
@@ -251,7 +251,7 @@ gpttype_adapter.o: gpttype_adapter.cpp
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 
 clean:
-	rm -vf *.o main quantize_llama quantize_gpt2 quantize_gptj quantize-stats perplexity embedding benchmark-q4_0-matmult main.exe quantize_llama.exe quantize_gptj.exe quantize_gpt2.exe koboldcpp.dll koboldcpp_openblas.dll koboldcpp_noavx2.dll koboldcpp_openblas_noavx2.dll koboldcpp_clblast.dll koboldcpp.so koboldcpp_openblas.so koboldcpp_noavx2.so koboldcpp_openblas_noavx2.so koboldcpp_clblast.so gptj.exe gpt2.exe
+	rm -vf *.o main quantize_llama quantize_gpt2 quantize_gptj quantize-stats perplexity embedding benchmark-q4_0-matmult main.exe quantize_llama.exe quantize_gptj.exe quantize_gpt2.exe koboldcpp.dll koboldcpp_openblas.dll koboldcpp_noavx2.dll koboldcpp_openblas_noavx2.dll koboldcpp_clblast.dll koboldcpp_cublas.dll koboldcpp.so koboldcpp_openblas.so koboldcpp_noavx2.so koboldcpp_openblas_noavx2.so koboldcpp_clblast.so koboldcpp_cublas.so gptj.exe gpt2.exe
 
 main: examples/main/main.cpp ggml.o llama.o common.o $(OBJS)
 	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
diff --git a/README.md b/README.md
index 5ec570b62..bc483b785 100644
--- a/README.md
+++ b/README.md
@@ -36,7 +36,8 @@ What does it mean? You get llama.cpp with a fancy UI, persistent stories, editin
 - You will have to compile your binaries from source. A makefile is provided, simply run `make`
 - If you want you can also link your own install of OpenBLAS manually with `make LLAMA_OPENBLAS=1`
 - Alternatively, if you want you can also link your own install of CLBlast manually with `make LLAMA_CLBLAST=1`, for this you will need to obtain and link OpenCL and CLBlast libraries.
-  - For Arch Linux: Install `cblas` and `openblas`. In the makefile, find the `ifdef LLAMA_OPENBLAS` conditional and add `-lcblas` to `LDFLAGS`.
+- For a full featured build, do `make LLAMA_OPENBLAS=1 LLAMA_CLBLAST=1`
+  - For Arch Linux: Install `cblas` and `openblas`. 
   - For Debian: Install `libclblast-dev` and `libopenblas-dev`.
 - After all binaries are built, you can run the python script with the command `koboldcpp.py [ggml_model.bin] [port]`
 
diff --git a/CL/Utils/Context.h b/include/CL/Utils/Context.h
similarity index 100%
rename from CL/Utils/Context.h
rename to include/CL/Utils/Context.h
diff --git a/CL/Utils/Context.hpp b/include/CL/Utils/Context.hpp
similarity index 100%
rename from CL/Utils/Context.hpp
rename to include/CL/Utils/Context.hpp
diff --git a/CL/Utils/Detail.hpp b/include/CL/Utils/Detail.hpp
similarity index 100%
rename from CL/Utils/Detail.hpp
rename to include/CL/Utils/Detail.hpp
diff --git a/CL/Utils/Device.hpp b/include/CL/Utils/Device.hpp
similarity index 100%
rename from CL/Utils/Device.hpp
rename to include/CL/Utils/Device.hpp
diff --git a/CL/Utils/Error.h b/include/CL/Utils/Error.h
similarity index 100%
rename from CL/Utils/Error.h
rename to include/CL/Utils/Error.h
diff --git a/CL/Utils/Error.hpp b/include/CL/Utils/Error.hpp
similarity index 100%
rename from CL/Utils/Error.hpp
rename to include/CL/Utils/Error.hpp
diff --git a/CL/Utils/ErrorCodes.h b/include/CL/Utils/ErrorCodes.h
similarity index 100%
rename from CL/Utils/ErrorCodes.h
rename to include/CL/Utils/ErrorCodes.h
diff --git a/CL/Utils/Event.h b/include/CL/Utils/Event.h
similarity index 100%
rename from CL/Utils/Event.h
rename to include/CL/Utils/Event.h
diff --git a/CL/Utils/Event.hpp b/include/CL/Utils/Event.hpp
similarity index 100%
rename from CL/Utils/Event.hpp
rename to include/CL/Utils/Event.hpp
diff --git a/CL/Utils/File.h b/include/CL/Utils/File.h
similarity index 100%
rename from CL/Utils/File.h
rename to include/CL/Utils/File.h
diff --git a/CL/Utils/File.hpp b/include/CL/Utils/File.hpp
similarity index 100%
rename from CL/Utils/File.hpp
rename to include/CL/Utils/File.hpp
diff --git a/CL/Utils/InteropContext.hpp b/include/CL/Utils/InteropContext.hpp
similarity index 100%
rename from CL/Utils/InteropContext.hpp
rename to include/CL/Utils/InteropContext.hpp
diff --git a/CL/Utils/OpenCLUtilsCpp_Export.h b/include/CL/Utils/OpenCLUtilsCpp_Export.h
similarity index 100%
rename from CL/Utils/OpenCLUtilsCpp_Export.h
rename to include/CL/Utils/OpenCLUtilsCpp_Export.h
diff --git a/CL/Utils/OpenCLUtils_Export.h b/include/CL/Utils/OpenCLUtils_Export.h
similarity index 100%
rename from CL/Utils/OpenCLUtils_Export.h
rename to include/CL/Utils/OpenCLUtils_Export.h
diff --git a/CL/Utils/Platform.hpp b/include/CL/Utils/Platform.hpp
similarity index 100%
rename from CL/Utils/Platform.hpp
rename to include/CL/Utils/Platform.hpp
diff --git a/CL/Utils/Utils.h b/include/CL/Utils/Utils.h
similarity index 100%
rename from CL/Utils/Utils.h
rename to include/CL/Utils/Utils.h
diff --git a/CL/Utils/Utils.hpp b/include/CL/Utils/Utils.hpp
similarity index 100%
rename from CL/Utils/Utils.hpp
rename to include/CL/Utils/Utils.hpp
diff --git a/CL/cl.h b/include/CL/cl.h
similarity index 100%
rename from CL/cl.h
rename to include/CL/cl.h
diff --git a/CL/cl2.hpp b/include/CL/cl2.hpp
similarity index 100%
rename from CL/cl2.hpp
rename to include/CL/cl2.hpp
diff --git a/CL/cl_d3d10.h b/include/CL/cl_d3d10.h
similarity index 100%
rename from CL/cl_d3d10.h
rename to include/CL/cl_d3d10.h
diff --git a/CL/cl_d3d11.h b/include/CL/cl_d3d11.h
similarity index 100%
rename from CL/cl_d3d11.h
rename to include/CL/cl_d3d11.h
diff --git a/CL/cl_dx9_media_sharing.h b/include/CL/cl_dx9_media_sharing.h
similarity index 100%
rename from CL/cl_dx9_media_sharing.h
rename to include/CL/cl_dx9_media_sharing.h
diff --git a/CL/cl_dx9_media_sharing_intel.h b/include/CL/cl_dx9_media_sharing_intel.h
similarity index 100%
rename from CL/cl_dx9_media_sharing_intel.h
rename to include/CL/cl_dx9_media_sharing_intel.h
diff --git a/CL/cl_egl.h b/include/CL/cl_egl.h
similarity index 100%
rename from CL/cl_egl.h
rename to include/CL/cl_egl.h
diff --git a/CL/cl_ext.h b/include/CL/cl_ext.h
similarity index 100%
rename from CL/cl_ext.h
rename to include/CL/cl_ext.h
diff --git a/CL/cl_ext_intel.h b/include/CL/cl_ext_intel.h
similarity index 100%
rename from CL/cl_ext_intel.h
rename to include/CL/cl_ext_intel.h
diff --git a/CL/cl_gl.h b/include/CL/cl_gl.h
similarity index 100%
rename from CL/cl_gl.h
rename to include/CL/cl_gl.h
diff --git a/CL/cl_gl_ext.h b/include/CL/cl_gl_ext.h
similarity index 100%
rename from CL/cl_gl_ext.h
rename to include/CL/cl_gl_ext.h
diff --git a/CL/cl_half.h b/include/CL/cl_half.h
similarity index 100%
rename from CL/cl_half.h
rename to include/CL/cl_half.h
diff --git a/CL/cl_icd.h b/include/CL/cl_icd.h
similarity index 100%
rename from CL/cl_icd.h
rename to include/CL/cl_icd.h
diff --git a/CL/cl_layer.h b/include/CL/cl_layer.h
similarity index 100%
rename from CL/cl_layer.h
rename to include/CL/cl_layer.h
diff --git a/CL/cl_platform.h b/include/CL/cl_platform.h
similarity index 100%
rename from CL/cl_platform.h
rename to include/CL/cl_platform.h
diff --git a/CL/cl_va_api_media_sharing_intel.h b/include/CL/cl_va_api_media_sharing_intel.h
similarity index 100%
rename from CL/cl_va_api_media_sharing_intel.h
rename to include/CL/cl_va_api_media_sharing_intel.h
diff --git a/CL/cl_version.h b/include/CL/cl_version.h
similarity index 100%
rename from CL/cl_version.h
rename to include/CL/cl_version.h
diff --git a/CL/license.txt b/include/CL/license.txt
similarity index 100%
rename from CL/license.txt
rename to include/CL/license.txt
diff --git a/CL/opencl.h b/include/CL/opencl.h
similarity index 100%
rename from CL/opencl.h
rename to include/CL/opencl.h
diff --git a/CL/opencl.hpp b/include/CL/opencl.hpp
similarity index 100%
rename from CL/opencl.hpp
rename to include/CL/opencl.hpp
diff --git a/cblas.h b/include/cblas.h
similarity index 99%
rename from cblas.h
rename to include/cblas.h
index fb8102294..48b47059d 100644
--- a/cblas.h
+++ b/include/cblas.h
@@ -4,7 +4,7 @@
 #define CBLAS_H
 
 #include <stddef.h>
-#include "openblas_config.h"
+#include <openblas_config.h>
 
 #ifdef __cplusplus
 extern "C" {
diff --git a/clblast_c.h b/include/clblast_c.h
similarity index 100%
rename from clblast_c.h
rename to include/clblast_c.h
diff --git a/openblas_config.h b/include/openblas_config.h
similarity index 100%
rename from openblas_config.h
rename to include/openblas_config.h