Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-02-02 15:02:47 +01:00
Code Issues Packages Projects Releases Wiki Activity

metal : use mm kernel only for quantum KV cache

Browse Source
This commit is contained in:
Georgi Gerganov 2023-12-05 18:14:04 +02:00
parent 903167a777
commit dd86df82e6
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ggml-metal.m
View File

@ -1129,7 +1129,7 @@ void ggml_metal_graph_compute(
!ggml_is_transposed(src1) &&
src1t == GGML_TYPE_F32 &&
ne00 % 32 == 0 && ne00 >= 64 &&
(ne11 > ne11_mm_min || ne12 > 1)) {
(ne11 > ne11_mm_min || (ggml_is_quantized(src0t) && ne12 > 1))) {
//printf("matrix: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
switch (src0->type) {
case GGML_TYPE_F32: [encoder setComputePipelineState:ctx->pipeline_mul_mm_f32_f32]; break;