updated to handle changes for clip_is_mrope

tools/mtmd/clip.cpp

@@ -4407,19 +4407,6 @@ bool clip_is_glm(const struct clip_ctx * ctx) {
     return ctx->proj_type() == PROJECTOR_TYPE_GLM_EDGE;
 }
 
-bool clip_is_mrope(const struct clip_ctx * ctx) { //kcpp: this was removed in https://github.com/ggml-org/llama.cpp/pull/18793 and moved to mtmd_decode_use_mrope
-    switch (ctx->proj_type()) {
-        case PROJECTOR_TYPE_QWEN2VL:
-        case PROJECTOR_TYPE_QWEN25VL:
-        case PROJECTOR_TYPE_QWEN3VL:
-        case PROJECTOR_TYPE_GLM4V:
-        case PROJECTOR_TYPE_PADDLEOCR:
-            return true;
-        default:
-            return false;
-    }
-}
-
 bool clip_is_llava(const struct clip_ctx * ctx) {
     return ctx->model.hparams.has_llava_projector;
 }

tools/mtmd/clip.h

@@ -113,7 +113,6 @@ 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_mrope(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

@@ -49,7 +49,7 @@ static bool encode_image_with_clip(clip_ctx * ctx_clip, int n_threads, struct cl
     return true;
 }
 
-bool llava_image_embed_make_with_clip_img(clip_ctx * ctx_clip, int n_threads, const clip_image_u8 * img, float ** image_embd_out, int * n_img_pos_out, int * nx_out, int * ny_out) {
+bool llava_image_embed_make_with_clip_img(clip_ctx * ctx_clip, int n_threads, const clip_image_u8 * img, float ** image_embd_out, int * n_img_pos_out, int * nx_out, int * ny_out, bool clip_is_mrope) {
     // Granite vision uses up to 10 patches + base patch
     int num_max_patches = 11;
     if (clip_is_minicpmv(ctx_clip)) {
@@ -65,7 +65,7 @@ bool llava_image_embed_make_with_clip_img(clip_ctx * ctx_clip, int n_threads, co
         return false;
     }
 
-    if (clip_is_mrope(ctx_clip)) {
+    if (clip_is_mrope) {
         // qwen2vl don't split image into chunks, so `num_max_patches` is not needed.
         //sometimes they resize the image LARGER than before (padding up), so we must account for that
         int max_nx = img->nx;

tools/mtmd/llava.h

@@ -30,7 +30,7 @@ struct llava_image_embed {
 struct mtmd_audio_mel;
 
 
-LLAVA_API bool llava_image_embed_make_with_clip_img(struct clip_ctx * ctx_clip, int n_threads, const struct clip_image_u8 * img, float ** image_embd_out, int * n_img_pos_out, int * nx_out, int * ny_out);
+LLAVA_API bool llava_image_embed_make_with_clip_img(struct clip_ctx * ctx_clip, int n_threads, const struct clip_image_u8 * img, float ** image_embd_out, int * n_img_pos_out, int * nx_out, int * ny_out, bool clip_is_mrope);
 
 LLAVA_API bool audio_embd_make_with_clip_img(clip_ctx * ctx_clip, int n_threads, const mtmd_audio_mel & mel_spec, float ** image_embd_out, int * n_img_pos_out);