Update app.py

app.py

@@ -62,10 +62,21 @@ def transcribe_audio(audio_path: str, whisper_model) -> str:  # Pass whisper_mod
 
 def generate_response(text: str, model_gemma, tokenizer_gemma, device) -> str: # Pass model and tokenizer
     try:
-        input_text = "Reapond to the users prompt: " + text
-        input = tokenizer_gemma(input_text, return_tensors="pt").to(device)
-        generated_output = model_gemma.generate(**input, max_length=MAX_GEMMA_LENGTH, early_stopping=True)
+        # Gemma 3 chat template format
+        messages = [{"role": "user", "content": text}]
+        input = tokenizer_gemma.apply_chat_template(messages, return_tensors="pt").to(device)
+        generation_config = GenerationConfig(
+            max_new_tokens=MAX_GEMMA_LENGTH,
+            early_stopping=True,
+        )
+
+        generated_output = model_gemma.generate(input, generation_config=generation_config)
         return tokenizer_gemma.decode(generated_output[0], skip_special_tokens=True)
+        
+        #input_text = "Reapond to the users prompt: " + text
+        #input = tokenizer_gemma(input_text, return_tensors="pt").to(device)
+        #generated_output = model_gemma.generate(**input, max_length=MAX_GEMMA_LENGTH, early_stopping=True)
+        #return tokenizer_gemma.decode(generated_output[0], skip_special_tokens=True)
     except Exception as e:
         logging.error(f"Gemma response generation error: {e}")
         return "I'm sorry, I encountered an error generating a response."
@@ -144,8 +155,8 @@ def _infer(user_audio, generator, whisper_model, tokenizer_gemma, model_gemma, d
              raise gr.Error(f"Sesame response generation error: {e}")
 
 
-        user_segment = Segment(speaker = SPEAKER_ID, text = 'User Audio', audio = load_audio(user_audio, generator)) #Pass Generator
-        ai_segment =  Segment(speaker = SPEAKER_ID, text = 'AI Audio', audio = ai_audio)
+        user_segment = Segment(speaker = 1, text = user_text, audio = load_audio(user_audio, generator)) #Pass Generator
+        ai_segment =  Segment(speaker = SPEAKER_ID, text = ai_text, audio = ai_audio)
         conversation_history.append(user_segment)
         conversation_history.append(ai_segment)