CUDA 12 support for the Conda Runtime (LostRuins#680)

* Remove libculibos dependency

This dependency is something that is used to build libcudart which we are also already targeting. The individual file is no longer being distributed with the CUDA12 conda devkit, so we can no longer target it directly. But because all its functionality is inside libcudart we also don't need it.

This commit removes the inclusion so that Koboldcpp can be compiled with CUDA12 as distributed by conda. I have tested this on the 1.57 release on CUDA11.5 and CUDA12.1.

* Cleanup version definitions

The package versions are already controlled by the label, we don't need to define it multiple times for it to work correctly. Removing the separate definitions allows people to easily change which version of CUDA they wish for their system.

Makefile

@@ -141,7 +141,7 @@ endif
 # it is recommended to use the CMAKE file to build for cublas if you can - will likely work better
 ifdef LLAMA_CUBLAS
 	CUBLAS_FLAGS = -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include
-	CUBLASLD_FLAGS = -lcuda -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L$(CUDA_PATH)/targets/x86_64-linux/lib -L/usr/local/cuda/targets/aarch64-linux/lib -L/usr/lib/wsl/lib
+	CUBLASLD_FLAGS = -lcuda -lcublas -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L$(CUDA_PATH)/targets/x86_64-linux/lib -L/usr/local/cuda/targets/aarch64-linux/lib -L/usr/lib/wsl/lib
 	CUBLAS_OBJS = ggml-cuda.o ggml_v3-cuda.o ggml_v2-cuda.o ggml_v2-cuda-legacy.o
 	NVCC      = nvcc
 	NVCCFLAGS = --forward-unknown-to-host-compiler -use_fast_math

environment.yaml

@@ -5,8 +5,8 @@ channels:
   - defaults
 dependencies:
   - python=3.8
-  - cuda-nvcc=11.5
-  - cuda-libraries-dev=11.5
+  - cuda-nvcc
+  - cuda-libraries-dev
   - cxx-compiler
   - gxx=10
   - pip