gemma4uv alloc fix

tools/mtmd/clip.cpp

@@ -5145,8 +5145,8 @@ bool clip_is_llava(const struct clip_ctx * ctx) {
     return ctx->model.hparams.has_llava_projector;
 }
 
-bool clip_is_gemma4(const struct clip_ctx * ctx) { //for kcpp use
-    return ctx->proj_type() == PROJECTOR_TYPE_GEMMA4V;
+bool clip_is_gemma4v(const struct clip_ctx * ctx) { //for kcpp use
+    return (ctx->proj_type() == PROJECTOR_TYPE_GEMMA4V || ctx->proj_type() == PROJECTOR_TYPE_GEMMA4UV);
 }
 
 bool clip_has_vision_encoder(const struct clip_ctx * ctx) {

tools/mtmd/clip.h

@@ -115,7 +115,7 @@ bool clip_image_batch_encode(struct clip_ctx * ctx, int n_threads, const struct
 
 int clip_is_minicpmv(const struct clip_ctx * ctx);
 bool clip_is_glm(const struct clip_ctx * ctx);
-bool clip_is_gemma4(const struct clip_ctx * ctx);
+bool clip_is_gemma4v(const struct clip_ctx * ctx);
 bool clip_is_llava(const struct clip_ctx * ctx);
 void set_clip_uses_gpu(bool usegpu);
 int clip_get_projector_type_ext(clip_ctx * ctx);

tools/mtmd/llava.cpp

@@ -79,7 +79,7 @@ bool llava_image_embed_make_with_clip_img(clip_ctx * ctx_clip, int n_threads, co
         }
         image_embd = (float *)malloc(clip_embd_nbytes_by_img(ctx_clip, max_nx, max_ny));
     }
-    else if(clip_is_gemma4(ctx_clip)) //gemma4 vision
+    else if(clip_is_gemma4v(ctx_clip)) //gemma4 vision
     {
         const size_t n_imgs = clip_image_f32_batch_n_images(preprocessed_img.get());
         clip_image_f32 * img_res = clip_image_f32_get_img(preprocessed_img.get(), 0);