ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-30 19:31:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fused rms_norm WIP

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-09-08 07:06:38 +02:00
parent 3c46f6f817
commit d2225010b9
2 changed files with 6 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
ggml/src/ggml-metal.m
View File

@@ -2616,6 +2616,7 @@ static enum ggml_status ggml_metal_graph_compute(
GGML_ASSERT(ne00 % 4 == 0); GGML_ASSERT(ne00 % 4 == 0);
GGML_ASSERT(ggml_is_contiguous_1(src0)); GGML_ASSERT(ggml_is_contiguous_1(src0));
GGML_ASSERT(src1->ne[0] == src0->ne[0]); GGML_ASSERT(src1->ne[0] == src0->ne[0]);
GGML_ASSERT(src1->type == GGML_TYPE_F32);
GGML_ASSERT(ggml_nrows(src1) == 1); GGML_ASSERT(ggml_nrows(src1) == 1);
float eps; float eps;

7
src/llama.cpp
View File

@@ -7988,9 +7988,12 @@ static struct ggml_tensor * llm_build_norm(
const llm_build_cb & cb, const llm_build_cb & cb,
int il, float scale_eps = 1) { int il, float scale_eps = 1) {
if (type == LLM_NORM_RMS && !mb) { if (type == LLM_NORM_RMS && mw) {
cur = ggml_fused_rms_norm(ctx, cur, mw, scale_eps * hparams.f_norm_rms_eps); cur = ggml_fused_rms_norm(ctx, cur, mw, scale_eps * hparams.f_norm_rms_eps);
cb(cur, "fused_norm", il); if (mb) {
cb(cur, "fused_norm", il);
cur = ggml_add(ctx, cur, mb);
}
return cur; return cur;
} }