CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-13 10:57:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

mtmd: fix patch_size initialized to random value in audio models (#17128)

Browse Source 
* mtmd: fix patch_size initialized to random value in audio models

* add default hparams
This commit is contained in:
Xuan-Son Nguyen
2025-11-10 11:41:05 +01:00
committed by GitHub
parent 9898b57cbe
commit 4b13a684c5
1 changed files with 10 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
tools/mtmd/clip.cpp
View File

@@ -160,13 +160,13 @@ enum patch_merge_type {
}; };
struct clip_hparams { struct clip_hparams {
int32_t image_size; int32_t image_size = 0;
int32_t patch_size; int32_t patch_size = 0;
int32_t n_embd; int32_t n_embd = 0;
int32_t n_ff; int32_t n_ff = 0;
int32_t projection_dim; int32_t projection_dim = 0;
int32_t n_head; int32_t n_head = 0;
int32_t n_layer; int32_t n_layer = 0;
// idefics3 // idefics3
int32_t image_longest_edge = 0; int32_t image_longest_edge = 0;
int32_t image_min_pixels = -1; int32_t image_min_pixels = -1;
@@ -2683,6 +2683,9 @@ struct clip_model_loader {
} }
} else if (is_audio) { } else if (is_audio) {
get_u32(KEY_A_NUM_MEL_BINS, hparams.n_mel_bins); get_u32(KEY_A_NUM_MEL_BINS, hparams.n_mel_bins);
// some hparams are unused, but still need to set to avoid issues
hparams.image_size = 0;
hparams.patch_size = 1;
} else { } else {
GGML_ASSERT(false && "unknown modality"); GGML_ASSERT(false && "unknown modality");