vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-09 19:46:11 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

graph : reduce topology branching (#18548)

Browse source
This commit is contained in:
Georgi Gerganov 2026-01-02 19:01:56 +02:00 • committed by GitHub
parent d84a6a98be
commit af1e8e1a6c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 14 additions and 20 deletions
Download patch file Download diff file Expand all files Collapse all files

8
src/models/gemma-embedding.cpp
View file

@ -1,7 +1,5 @@
#include "models.h"
llm_build_gemma_embedding::llm_build_gemma_embedding(const llama_model & model, const llm_graph_params & params) :
llm_graph_context(params) {
const int64_t n_embd_head = hparams.n_embd_head_k;
@ -12,10 +10,8 @@ llm_build_gemma_embedding::llm_build_gemma_embedding(const llama_model & model,
inpL = build_inp_embd(model.tok_embd);
// important: do not normalize weights for raw embeddings input (i.e. encoded image emdeddings)
if (ubatch.token) {
inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
cb(inpL, "inp_scaled", -1);
}
inpL = ggml_scale(ctx0, inpL, ubatch.token ? sqrtf(n_embd) : 1.0f);
cb(inpL, "inp_scaled", -1);
// inp_pos - contains the positions
ggml_tensor * inp_pos = build_inp_pos();