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

cuda : increase max block size to 1024

Browse source
This commit is contained in:
Georgi Gerganov 2023-11-30 20:40:47 +02:00
parent 62532c05aa
commit 6b86bcffac
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml-cuda.cu
View file

@ -443,7 +443,7 @@ static_assert(sizeof(block_q6_K) == sizeof(ggml_fp16_t) + 13*QK_K/16, "wrong q6_
#define CUDA_SCALE_BLOCK_SIZE 256
#define CUDA_CLAMP_BLOCK_SIZE 256
#define CUDA_ROPE_BLOCK_SIZE 256
#define CUDA_SOFT_MAX_BLOCK_SIZE 512
#define CUDA_SOFT_MAX_BLOCK_SIZE 1024
#define CUDA_ALIBI_BLOCK_SIZE 32
#define CUDA_DIAG_MASK_INF_BLOCK_SIZE 32
#define CUDA_QUANTIZE_BLOCK_SIZE 256