temp

Author: Wovchena

src/cpp/src/vision_encoder.cpp

@@ -10,6 +10,14 @@
 using namespace ov::genai;
 
 namespace {
+/// @brief A pair describing image size.
+struct HeightWidth {
+    /// @brief Height of a corresponding image.
+    size_t height;
+    /// @brief Width of a corresponding image.
+    size_t width;
+};
+
 int ensure_divide(int length, int patch_size) {
     return std::max(static_cast<int>(std::round(static_cast<float>(length) / patch_size) * patch_size), patch_size);
 }

src/cpp/src/vlm_pipeline.cpp

@@ -29,6 +29,15 @@ struct Args {
     float repeat_penalty = 1.0f;
 };
 
+/// @brief A pair describing image size.
+struct HeightWidth {
+    /// @brief Height of a corresponding image.
+    size_t height;
+    /// @brief Width of a corresponding image.
+    size_t width;
+};
+
+
 int64_t get_out_token_id(const std::vector<int>& input_ids, float* logits, size_t vocab_size, Args args) {
     int64_t out_token;