vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

update makefile and cuda,fix some issue

Browse source
This commit is contained in:
FantasyGmm 2023-12-22 11:28:19 +08:00
parent e6d10852d5
commit 525ec7e475
2 changed files with 6 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

6
Makefile
View file

@ -366,10 +366,14 @@ ifdef LLAMA_BLIS
endif # LLAMA_BLIS
ifdef LLAMA_CUBLAS
MK_CPPFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -L$(CUDA_PATH)/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include
MK_CPPFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include
MK_LDFLAGS += -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
OBJS += ggml-cuda.o
ifdef JETSON_EOL_MODULE_DETECT
MK_NVCCFLAGS = -use_fast_math
else
MK_NVCCFLAGS = --forward-unknown-to-host-compiler -use_fast_math
endif # JETSON_EOL_MODULE_DETECT
ifdef LLAMA_DEBUG
MK_NVCCFLAGS += -lineinfo

2
ggml-cuda.cu
View file

@ -8403,7 +8403,7 @@ static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const
half * dst_f16 = nullptr;
char * dst_t = nullptr;
cudaDataType_t cu_compute_type = CUBLAS_COMPUTE_16F;
cublasComputeType_t cu_compute_type = CUBLAS_COMPUTE_16F;
cudaDataType_t cu_data_type = CUDA_R_16F;
// dst strides