CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml : change ggml_scale to take a float instead of tensor (#4573)

Browse Source 
* ggml : change ggml_scale to take a float instead of tensor

* ggml : fix CPU implementation

* tests : fix test-grad0

ggml-ci
This commit is contained in:
Georgi Gerganov
2023-12-21 23:20:49 +02:00
committed by GitHub
parent 769a7bc85e
commit afefa319f1
12 changed files with 82 additions and 205 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ggml-metal.m
View File

@@ -1293,7 +1293,7 @@ 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);
@@ -1304,8 +1304,8 @@ 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)];