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

metal : do not use mul_mm kernels when ne00 < 64

Browse source
This commit is contained in:
Georgi Gerganov 2023-10-08 12:05:15 +03:00
parent 94e502dfb7
commit fdd5ad9a54
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-metal.m
View file

@ -1040,7 +1040,7 @@ void ggml_metal_graph_compute(
!ggml_is_transposed(src0) && !ggml_is_transposed(src0) &&
!ggml_is_transposed(src1) && !ggml_is_transposed(src1) &&
src1t == GGML_TYPE_F32 && src1t == GGML_TYPE_F32 &&
ne00 % 32 == 0 && ne00 % 32 == 0 && ne00 >= 64 &&
ne11 > ne11_mm_min) { ne11 > ne11_mm_min) {
//printf("matrix: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12); //printf("matrix: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
switch (src0->type) { switch (src0->type) {