Skip to content

Commit e3ba22d

Browse files
abetlenngxson
andauthored
fix(mtmd): handle Gemma 4 audio projector embedding size (#24091)
* mtmd: handle Gemma 4 audio projector embedding size * rm projection_dim from clip_n_mmproj_embd --------- Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
1 parent 6ddc943 commit e3ba22d

1 file changed

Lines changed: 2 additions & 2 deletions

File tree

tools/mtmd/clip.cpp

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -4347,6 +4347,8 @@ int clip_n_mmproj_embd(const struct clip_ctx * ctx) {
43474347
return ctx->model.mm_input_proj_w->ne[0];
43484348
case PROJECTOR_TYPE_GEMMA4V:
43494349
case PROJECTOR_TYPE_GEMMA4UV:
4350+
case PROJECTOR_TYPE_GEMMA4A:
4351+
case PROJECTOR_TYPE_GEMMA4UA:
43504352
return ctx->model.mm_input_proj_w->ne[1];
43514353
case PROJECTOR_TYPE_IDEFICS3:
43524354
return ctx->model.mm_fc_w->ne[1];
@@ -4381,8 +4383,6 @@ int clip_n_mmproj_embd(const struct clip_ctx * ctx) {
43814383
return ctx->model.mm_fc_w->ne[1];
43824384
case PROJECTOR_TYPE_LFM2A:
43834385
return ctx->model.position_embeddings->ne[0];
4384-
case PROJECTOR_TYPE_GEMMA4UA:
4385-
return ctx->model.hparams.projection_dim;
43864386
case PROJECTOR_TYPE_GRANITE_SPEECH:
43874387
return ctx->model.qf_proj_linear_w->ne[1];
43884388
case PROJECTOR_TYPE_GLM4V:

0 commit comments

Comments
 (0)