Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

convert : add --fuse-qkv flag to fuse Q/K/V into QKV during HF-to-GGUF conversion (#22780)

Browse Source
This commit is contained in:
PikaPikachu
2026-09-07 00:47:05 +08:00
committed by GitHub
parent 5fdfa62829
commit 465e49b9ce
20 changed files with 396 additions and 125 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/olmo2.cpp
+5 -6
View File
@@ -93,14 +93,13 @@ llama_model_olmo2::graph<iswa>::graph(const llama_model & model, const llm_graph
// self_attention
{
// compute Q and K and RoPE them
ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
auto [Qcur, Kcur, Vcur] = build_qkv(model.layers[il], cur,
n_embd_head, n_head,
n_embd_head, n_head_kv,
n_embd_head, n_head_kv,
il, false);
cb(Qcur, "Qcur", il);
ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
cb(Kcur, "Kcur", il);
ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
cb(Vcur, "Vcur", il);
Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL,