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

Updated Makefile

Browse source
This commit is contained in:
JohannesGaessler 2023-07-29 08:03:30 +02:00
parent 3c09e11c97
commit 038ed63195
2 changed files with 7 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

5
Makefile
View file

@ -220,6 +220,11 @@ else ifdef LLAMA_CUDA_DMMV_Y
else
NVCCFLAGS += -DGGML_CUDA_MMV_Y=1
endif # LLAMA_CUDA_MMV_Y
ifdef LLAMA_CUDA_MMQ_Y
NVCCFLAGS += -DGGML_CUDA_MMQ_Y=$(LLAMA_CUDA_MMQ_Y)
else
NVCCFLAGS += -DGGML_CUDA_MMQ_Y=64
endif # LLAMA_CUDA_MMQ_Y
ifdef LLAMA_CUDA_DMMV_F16
NVCCFLAGS += -DGGML_CUDA_DMMV_F16
endif # LLAMA_CUDA_DMMV_F16

4
ggml-cuda.cu
View file

@ -263,8 +263,8 @@ static_assert(sizeof(block_q6_K) == sizeof(ggml_fp16_t) + 13*QK_K/16, "wrong q6_
#define CUDA_DEQUANTIZE_BLOCK_SIZE 256
#ifndef GGML_CUDA_MMQ_Y
#define GGML_CUDA_MMQ_Y 128
#endif
#define GGML_CUDA_MMQ_Y 64
#endif // GGML_CUDA_MMQ_Y
// dmmv = dequantize_mul_mat_vec
#ifndef GGML_CUDA_DMMV_X