kanshan/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-11 05:30:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

graph : reduce topology branching

Browse Source
This commit is contained in:
Georgi Gerganov
2026-01-02 15:35:26 +02:00
parent c6f0e832da
commit 4ed59dc2c7
4 changed files with 14 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/gemma3.cpp
+3 -4
View File
@@ -10,10 +10,9 @@ llm_build_gemma3<iswa>::llm_build_gemma3(const llama_model & model, const llm_gr
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();