Fix mobile rendering and stabilize output display

Implemented st.empty() placeholders and unique session keys to ensure output persistence across mobile browsers. Refactored the composition column to prevent UI freezing during long-running model inference.

app.py

@@ -76,15 +76,19 @@ with col1:
 
 with col2:
     st.subheader("Output")
+    
+    # Create a persistent placeholder for the output
+    output_placeholder = st.empty()
+    
     if generate_btn:
         with st.spinner("Model is writing..."):
-            # Load Engine (This uses the @st.cache_resource to avoid repeated loading)
+            # Load Engine
             model, tokenizer = load_studio_engine()
             
-            # Build Inference Prompt using the modular data script
+            # Build Inference Prompt
             prompt = build_inference_prompt(genre, artist, title)
             
-            # Generate using the universal engine in metrics script
+            # Generate Lyrics
             raw_output = execute_generation(
                 model, tokenizer, prompt, 
                 max_tokens=token_limit, 
@@ -92,7 +96,19 @@ with col2:
                 do_sample=True
             )
             
-            # Post-process and display
+            # Post-process formatting
             clean_lyrics = format_lyrics(raw_output)
-            st.text_area("Final Draft", clean_lyrics, height=400)
-            st.download_button("Export as TXT", clean_lyrics, file_name=f"{title}_lyrics.txt")
+            
+            # Update the placeholder specifically
+            output_placeholder.text_area(
+                "Final Draft", 
+                clean_lyrics, 
+                height=400, 
+                key="lyrics_output"    # adding a key helps mobile state persistence
+            )
+            
+            st.download_button(
+                "Export as TXT", 
+                clean_lyrics, 
+                file_name=f"{title}_lyrics.txt"
+            )