diff --git a/examples/llava/clip.cpp b/examples/llava/clip.cpp index d1e1eec1e..d8022edb4 100644 --- a/examples/llava/clip.cpp +++ b/examples/llava/clip.cpp @@ -2763,6 +2763,10 @@ int clip_n_patches_by_img(const struct clip_ctx * ctx, struct clip_image_f32 * i const auto & params = ctx->vision_model.hparams; int n_patches = (params.image_size / params.patch_size) * (params.image_size / params.patch_size); + if (ctx->proj_type == PROJECTOR_TYPE_GEMMA3) + { + n_patches = 256; //kcpp hardcode gemma3 vision to 256 size + } if (ctx->proj_type == PROJECTOR_TYPE_LDP || ctx->proj_type == PROJECTOR_TYPE_LDPV2 || ctx->proj_type == PROJECTOR_TYPE_GLM_EDGE) { n_patches /= 4;