Public/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-19 09:15:04 +02:00
Code Issues Packages Projects Releases Wiki Activity

mtmd: fix longest_edge ignoring min/max pixels

Browse Source
This commit is contained in:
Xuan Son Nguyen
2026-08-05 15:56:21 +02:00
parent cd0fa6051a
commit f1dffb70d9
3 changed files with 65 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/mtmd/clip.cpp
+3
View File
@@ -1433,6 +1433,7 @@ struct clip_model_loader {
// use default llava-uhd preprocessing params
get_u32(KEY_PROJ_SCALE_FACTOR, hparams.n_merge, false);
get_u32(KEY_PREPROC_IMAGE_SIZE, hparams.image_longest_edge, false);
hparams.set_limit_image_tokens();
} break;
case PROJECTOR_TYPE_LFM2:
{
@@ -1470,6 +1471,7 @@ struct clip_model_loader {
get_u32(KEY_SPATIAL_MERGE_SIZE, hparams.n_merge, false);
hparams.image_longest_edge = hparams.image_size;
get_u32(KEY_PREPROC_IMAGE_SIZE, hparams.image_longest_edge, false);
hparams.set_limit_image_tokens();
hparams.set_warmup_n_tokens(256); // avoid OOM on warmup
} break;
case PROJECTOR_TYPE_DOTS_OCR:
@@ -1594,6 +1596,7 @@ struct clip_model_loader {
if (hparams.image_longest_edge == 0) {
hparams.image_longest_edge = 3024;
}
hparams.set_limit_image_tokens();
hparams.warmup_image_size = hparams.image_size;
} break;
case PROJECTOR_TYPE_YOUTUVL: