happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

refactor: merge VAETKI positions case with QWEN2VL

This commit is contained in:
suhyun-hwang 2026-01-10 20:35:39 +09:00
parent d8e8b77c44
commit c5e9eac8c5
1 changed files with 4 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
tools/mtmd/clip.cpp
View File

@ -3499,11 +3499,14 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
case PROJECTOR_TYPE_QWEN2VL:
case PROJECTOR_TYPE_QWEN3VL:
case PROJECTOR_TYPE_GLM4V:
case PROJECTOR_TYPE_VAETKI:
{
const int merge_ratio = hparams.n_merge;
const int pw = image_size_width / patch_size;
const int ph = image_size_height / patch_size;
std::vector<int> positions(n_pos * 4);
const int pos_size = (model.proj_type == PROJECTOR_TYPE_VAETKI) ? num_patches : n_pos;
std::vector<int> positions(pos_size * 4);
int ptr = 0;
for (int y = 0; y < ph; y += merge_ratio) {
for (int x = 0; x < pw; x += merge_ratio) {
@ -3519,31 +3522,6 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
}
}
set_input_i32("positions", positions);
} break;
case PROJECTOR_TYPE_VAETKI:
{
const int merge_ratio = 2;
const int ipw = image_size_width / patch_size;
const int iph = image_size_height / patch_size;
std::vector<int> positions(num_patches * 4);
int ptr = 0;
for (int y = 0; y < iph; y += merge_ratio) {
for (int x = 0; x < ipw; x += merge_ratio) {
for (int dy = 0; dy < 2; dy++) {
for (int dx = 0; dx < 2; dx++) {
positions[ ptr] = y + dy;
positions[ num_patches + ptr] = x + dx;
positions[2 * num_patches + ptr] = y + dy;
positions[3 * num_patches + ptr] = x + dx;
ptr++;
}
}
}
}
set_input_i32("positions", positions);
} break;
case PROJECTOR_TYPE_QWEN25VL: