Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 14:20:31 +01:00
Code Issues Packages Projects Releases Wiki Activity

llama : de-duplicate deepseek2 norm

Browse Source
This commit is contained in:
Georgi Gerganov 2024-07-15 14:10:39 +03:00
parent bda62d7999
commit 3dfda05956
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 5 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/llama.cpp
View File

@ -12873,12 +12873,12 @@ struct llm_build_context {
struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
cb(ffn_inp, "ffn_inp", il);
if ((uint32_t) il < hparams.n_layer_dense_lead) {
cur = llm_build_norm(ctx0, ffn_inp, hparams,
model.layers[il].ffn_norm, NULL,
LLM_NORM_RMS, cb, il);
cb(cur, "ffn_norm", il);
if ((uint32_t) il < hparams.n_layer_dense_lead) {
cur = llm_build_ffn(ctx0, cur,
model.layers[il].ffn_up, NULL, NULL,
model.layers[il].ffn_gate, NULL, NULL,
@ -12888,11 +12888,6 @@ struct llm_build_context {
cb(cur, "ffn_out", il);
} else {
// MoE branch
cur = llm_build_norm(ctx0, ffn_inp, hparams,
model.layers[il].ffn_norm, NULL,
LLM_NORM_RMS, cb, il);
cb(cur, "ffn_norm", il);
ggml_tensor * moe_out =
llm_build_moe_ffn(ctx0, cur,
model.layers[il].ffn_gate_inp,