tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-18 05:56:00 -04:00
Code Issues Packages Projects Releases Wiki Activity

ggml : implement GEGLU_ERF and GEGLU_QUICK ops (#14445)

Browse Source
This commit is contained in:
Sigbjørn Skjæret
2025-07-03 23:07:22 +02:00
committed by GitHub
parent bee28421be
commit 28657a8229
20 changed files with 789 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
ggml/src/ggml.c
View File

@@ -1140,9 +1140,11 @@ static const char * GGML_GLU_OP_NAME[GGML_GLU_OP_COUNT] = {
"REGLU",
"GEGLU",
"SWIGLU",
"GEGLU_ERF",
"GEGLU_QUICK",
};
static_assert(GGML_GLU_OP_COUNT == 3, "GGML_GLU_OP_COUNT != 3");
static_assert(GGML_GLU_OP_COUNT == 5, "GGML_GLU_OP_COUNT != 5");
static_assert(sizeof(struct ggml_object)%GGML_MEM_ALIGN == 0, "ggml_object size must be a multiple of GGML_MEM_ALIGN");
@@ -2768,6 +2770,48 @@ struct ggml_tensor * ggml_swiglu_split(
return ggml_glu_impl(ctx, a, b, GGML_GLU_OP_SWIGLU, false);
}
// ggml_geglu_erf
struct ggml_tensor * ggml_geglu_erf(
struct ggml_context * ctx,
struct ggml_tensor * a) {
return ggml_glu_impl(ctx, a, NULL, GGML_GLU_OP_GEGLU_ERF, false);
}
struct ggml_tensor * ggml_geglu_erf_swapped(
struct ggml_context * ctx,
struct ggml_tensor * a) {
return ggml_glu_impl(ctx, a, NULL, GGML_GLU_OP_GEGLU_ERF, true);
}
struct ggml_tensor * ggml_geglu_erf_split(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b) {
return ggml_glu_impl(ctx, a, b, GGML_GLU_OP_GEGLU_ERF, false);
}
// ggml_geglu_quick
struct ggml_tensor * ggml_geglu_quick(
struct ggml_context * ctx,
struct ggml_tensor * a) {
return ggml_glu_impl(ctx, a, NULL, GGML_GLU_OP_GEGLU_QUICK, false);
}
struct ggml_tensor * ggml_geglu_quick_swapped(
struct ggml_context * ctx,
struct ggml_tensor * a) {
return ggml_glu_impl(ctx, a, NULL, GGML_GLU_OP_GEGLU_QUICK, true);
}
struct ggml_tensor * ggml_geglu_quick_split(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b) {
return ggml_glu_impl(ctx, a, b, GGML_GLU_OP_GEGLU_QUICK, false);
}
// ggml_norm
static struct ggml_tensor * ggml_norm_impl(