Update app.py

app.py

@@ -18,46 +18,24 @@ generation_config_multi = model_multi.default_generation_config
 # ---------------------------------
 # MULTI-TURN INFERENCE FUNCTION
 # ---------------------------------
-# def multi_turn_chat(user_input, audio_file, history, current_audio):
-#     try:
-#         if audio_file is not None:
-#             current_audio = audio_file  # Update state if a new file is uploaded
-
-#         if current_audio is None:
-#             return history + [("System", "❌ Please upload an audio file before chatting.")], history, current_audio
-
-#         sound = llava.Sound(current_audio)
-#         prompt = f"<sound>\n{user_input}"
-
-#         response = model_multi.generate_content([sound, prompt], generation_config=generation_config_multi)
-
-#         history.append((user_input, response))
-#         return history, history, current_audio
-#     except Exception as e:
-#         history.append((user_input, f"❌ Error: {str(e)}"))
-#         return history, history, current_audio
-
-def multi_turn_chat(user_input, audio_file, history, audio_history):
+def multi_turn_chat(user_input, audio_file, history, current_audio):
     try:
         if audio_file is not None:
-            audio_history.append(audio_file)  # Append new audio to the list
-
-        if not audio_history:
-            return history + [("System", "❌ Please upload an audio file before chatting.")], history, audio_history
+            current_audio = audio_file  # Update state if a new file is uploaded
 
-        # Create list of llava.Sound objects for each audio in history
-        audio_sounds = [llava.Sound(audio) for audio in audio_history]
+        if current_audio is None:
+            return history + [("System", "❌ Please upload an audio file before chatting.")], history, current_audio
 
-        # Add the user prompt after all audio sounds
+        sound = llava.Sound(current_audio)
         prompt = f"<sound>\n{user_input}"
-        response = model_multi.generate_content(audio_sounds + [prompt], generation_config=generation_config_multi)
 
-        history.append((user_input, response))
-        return history, history, audio_history
+        response = model_multi.generate_content([sound, prompt], generation_config=generation_config_multi)
 
+        history.append((user_input, response))
+        return history, history, current_audio
     except Exception as e:
         history.append((user_input, f"❌ Error: {str(e)}"))
-        return history, history, audio_history
+        return history, history, current_audio
 
 
 def speech_prompt_infer(audio_prompt_file):
@@ -142,16 +120,13 @@ with gr.Blocks(css="""
             user_input_multi = gr.Textbox(label="Your message", placeholder="Ask a question about the audio...", lines=8)
             btn_multi = gr.Button("Send")
             history_state = gr.State([])           # Chat history
-            # current_audio_state = gr.State(None)   # Most recent audio file path
-            audio_history_state = gr.State([])  # List of audio file paths
+            current_audio_state = gr.State(None)   # Most recent audio file path
 
 
             btn_multi.click(
                 fn=multi_turn_chat,
-                inputs=[user_input_multi, audio_input_multi, history_state, audio_history_state],
-                outputs=[chatbot, history_state, audio_history_state]
-                # inputs=[user_input_multi, audio_input_multi, history_state, current_audio_state],
-                # outputs=[chatbot, history_state, current_audio_state]
+                inputs=[user_input_multi, audio_input_multi, history_state, current_audio_state],
+                outputs=[chatbot, history_state, current_audio_state]
             )
             gr.Examples(
                 examples=[
@@ -207,7 +182,7 @@ To enable these capabilities, we propose several large-scale training datasets c
 
 💡 Audio Flamingo 3 has strong audio, music and speech understanding capabilities.
 
-💡 Audio Flamingo 3 supports on-demand thinking for chain-of-though reasoning.
+💡 Audio Flamingo 3 supports on-demand thinking for chain-of-thought reasoning.
 
 💡 Audio Flamingo 3 supports long audio and speech understanding for audios up to 10 minutes.