provide workaround for export openai whisper models (#1301)

Author: eaidova

optimum/exporters/openvino/convert.py

@@ -723,6 +723,13 @@ def export_from_model(
                     setattr(model.generation_config, param_name, param_value)
                     setattr(model.config, param_name, None)
 
+        # workaround for https://github.com/huggingface/transformers/issues/37172
+        if is_transformers_version(">=", "4.50.0") and model_type == "whisper":
+            if hasattr(model.config, "forced_decoder_ids"):
+                model.config.forced_decoder_ids = None
+            if hasattr(model, "generation_config") and hasattr(model.generation_config, "forced_decoder_ids"):
+                model.generation_config.input_ids = model.generation_config.forced_decoder_ids
+                model.generation_config.forced_decoder_ids = None
         # Saving the model config and preprocessor as this is needed sometimes.
         save_config(model.config, output)
         generation_config = getattr(model, "generation_config", None)