Fix hidden state size for StaticWhisperPipeline

src/cpp/src/whisper_pipeline_static.cpp

@@ -397,7 +397,17 @@ void add_attention_mask_input(std::shared_ptr<ov::Model> model) {
     pm.run_passes(model);
 }
 
-void reshape_to_static(std::shared_ptr<ov::Model> model, const uint32_t input_size, const uint32_t kvcache_size) {
+
+uint32_t get_encoder_hidden_state_size(const std::shared_ptr<ov::Model>& encoder) {
+    for (auto output : encoder->outputs()) {
+        if (output.get_any_name() == "last_hidden_state") {
+            return output.get_partial_shape()[1].get_length();
+        }
+    }
+    return 0;
+}
+
+void reshape_to_static(std::shared_ptr<ov::Model> model, const uint32_t input_size, const uint32_t kvcache_size, const uint32_t lhstate_size) {
     std::map<std::string, ov::PartialShape> new_shapes;
     for (auto input : model->inputs()) {
         const auto& input_name = input.get_any_name();
@@ -414,14 +424,14 @@ void reshape_to_static(std::shared_ptr<ov::Model> model, const uint32_t input_si
             const auto& partial_shape = input.get_partial_shape();
             new_shape = partial_shape;
             new_shape[0] = 1;     // batch_dim
-            new_shape[1] = 1500;  // FIXME: is it got from encoder output{'last_hidden_state'}
+            new_shape[1] = lhstate_size;  // from encoder output{'last_hidden_state'}
         } else if (input_name.find("past_key_values") != std::string::npos) {
             const auto& partial_shape = input.get_partial_shape();
             new_shape = partial_shape;
             new_shape[0] = 1;  // Use batch dim here
             new_shape[2] = input_name.find(".decoder") != std::string::npos
                                ? kvcache_size - input_size // kv_size for decoder
-                               : 1500;  // for encoder
+                               : lhstate_size;  // hidden state size for encoder
         }
         new_shapes.emplace(input_name, new_shape);
     }
@@ -538,8 +548,10 @@ WhisperPipeline::StaticWhisperPipeline::StaticWhisperPipeline(const std::filesys
     size_t max_sequence_length = 448;
 
     reshape_to_static_encoder(encoder_model);
-    reshape_to_static(decoder_model, 4, 4);
-    reshape_to_static(decoder_with_past_model, 1, max_sequence_length);
+
+    size_t last_hidden_state_size = get_encoder_hidden_state_size(encoder_model);
+    reshape_to_static(decoder_model, 4, 4, last_hidden_state_size);
+    reshape_to_static(decoder_with_past_model, 1, max_sequence_length, last_hidden_state_size);
 
     // Replace KV-tensors for the entire cache to tensors only for new token
     decoder_with_past_model = redirect_new_kv_to_output(decoder_with_past_model);