ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-18 20:30:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

Adding fused y*unary(x) op

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-09-30 08:29:34 +03:00
parent cce49832c1
commit 6ef4f28aae
3 changed files with 144 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/llama.cpp
View File

@@ -8083,6 +8083,13 @@ static struct ggml_tensor * llm_build_ffn(
cur = tmp;
}
if (type_gate == LLM_FFN_PAR &&
(type_op == LLM_FFN_SILU || type_op == LLM_FFN_RELU || (type_op == LLM_FFN_GELU && !act_scales))) {
cur = ggml_fused_mul_unary(ctx, cur, tmp, type_op == LLM_FFN_SILU ? GGML_UNARY_OP_SILU :
type_op == LLM_FFN_RELU ? GGML_UNARY_OP_RELU : GGML_UNARY_OP_GELU);
}
else {
switch (type_op) {
case LLM_FFN_SILU:
{
@@ -8122,6 +8129,7 @@ static struct ggml_tensor * llm_build_ffn(
cur = ggml_mul(ctx, cur, tmp);
cb(cur, "ffn_gate_par", il);
}
}
if (down) {
cur = llm_build_lora_mm(lctx, ctx, down, cur);