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

ggml : remove GGML_KQ_MASK_PAD constant (#17910)

Browse source 
* ggml : remove GGML_KQ_MASK_PAD constant

* cont : remove comment
This commit is contained in:
Georgi Gerganov 2025-12-10 20:53:16 +02:00 • committed by GitHub
parent 4df6e859e9
commit 4dff236a52
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 19 additions and 36 deletions
Download patch file Download diff file Expand all files Collapse all files

6
tools/mtmd/clip.cpp
View file

@ -775,10 +775,6 @@ struct clip_graph {
// if flash attn is used, we need to pad the mask and cast to f16
if (ctx->flash_attn_type == CLIP_FLASH_ATTN_TYPE_ENABLED) {
int n_pad = GGML_PAD(window_mask->ne[1], GGML_KQ_MASK_PAD) - window_mask->ne[1];
if (n_pad > 0) {
window_mask = ggml_pad(ctx0, window_mask, 0, n_pad, 0, 0);
}
window_mask = ggml_cast(ctx0, window_mask, GGML_TYPE_F16);
}
@ -791,7 +787,7 @@ struct clip_graph {
// loop over layers
for (int il = 0; il < n_layer; il++) {
auto & layer = model.layers[il];
const auto & layer = model.layers[il];
const bool full_attn = use_window_attn ? (il + 1) % n_wa_pattern == 0 : true;
ggml_tensor * cur = inpL; // inpL = residual, cur = hidden_states