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

Faster Q4_K on Metal (#2290)

Browse source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow 2023-07-20 15:18:43 +03:00 • committed by GitHub
parent fff0e0eafe
commit 785829dfe8
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 160 additions and 109 deletions
Download patch file Download diff file Expand all files Collapse all files

7
ggml-metal.m
View file

@ -694,8 +694,8 @@ void ggml_metal_graph_compute(
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
nth0 = 4;
nth1 = 16;
nth0 = 2;
nth1 = 32;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_K_f32];
} break;
case GGML_TYPE_Q5_K:
@ -739,7 +739,8 @@ void ggml_metal_graph_compute(
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:13];
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:14];
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1) {
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1 ||
src0t == GGML_TYPE_Q4_K) {
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7) / 8, ne11, 1) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}
else if (src0t == GGML_TYPE_Q2_K ||