thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

mtmd: add clip_graph::build_mm() (#20751)

Browse source 
* clip: add build_mm()

* apply to all models

* add TODO for bias overload
This commit is contained in:
Xuan-Son Nguyen 2026-03-19 13:11:39 +01:00 • committed by GitHub
parent cd708db0cc
commit 1e64534570
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
15 changed files with 75 additions and 66 deletions
Download patch file Download diff file Expand all files Collapse all files

8
tools/mtmd/models/youtuvl.cpp
View file

@ -43,7 +43,7 @@ ggml_cgraph * clip_graph_youtuvl::build() {
ctx0, inp,
3*patch_size* patch_size, Hm * Wm * m * m, 1);
}
inp = ggml_mul_mat(ctx0, model.patch_embeddings_0, inp);
inp = build_mm(model.patch_embeddings_0, inp);
if (model.patch_bias) {
inp = ggml_add(ctx0, inp, model.patch_bias);
@ -97,11 +97,11 @@ ggml_cgraph * clip_graph_youtuvl::build() {
// self-attention
{
ggml_tensor * Qcur = ggml_add(ctx0,
ggml_mul_mat(ctx0, layer.q_w, cur), layer.q_b);
build_mm(layer.q_w, cur), layer.q_b);
ggml_tensor * Kcur = ggml_add(ctx0,
ggml_mul_mat(ctx0, layer.k_w, cur), layer.k_b);
build_mm(layer.k_w, cur), layer.k_b);
ggml_tensor * Vcur = ggml_add(ctx0,
ggml_mul_mat(ctx0, layer.v_w, cur), layer.v_b);
build_mm(layer.v_w, cur), layer.v_b);
Qcur = ggml_reshape_3d(ctx0, Qcur, d_head, n_head, n_patches);
Kcur = ggml_reshape_3d(ctx0, Kcur, d_head, n_head, n_patches);