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

ggml : change ggml_scale to take a float instead of tensor

Browse source
This commit is contained in:
Georgi Gerganov 2023-12-21 20:50:24 +02:00
parent 8fe03ffdda
commit 199f6bdc46
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
10 changed files with 68 additions and 186 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml-metal.m
View file

@ -1261,7 +1261,7 @@ void ggml_metal_graph_compute(
{
GGML_ASSERT(ggml_is_contiguous(src0));
const float scale = *(const float *) src1->data;
const float scale = *(const float *) dst->op_params;
int64_t n = ggml_nelements(dst);
@ -1272,8 +1272,8 @@ void ggml_metal_graph_compute(
[encoder setComputePipelineState:ctx->pipeline_scale];
}
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&scale length:sizeof(scale) atIndex:2];
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];