Update app.py

app.py

@@ -43,8 +43,8 @@ This demo allows you to have a conversation with Sesame CSM 1B, leveraging Whisp
 
 # --- Constants ---  (Constants can stay outside)
 SPEAKER_ID = 0
-MAX_CONTEXT_SEGMENTS = 1
-MAX_GEMMA_LENGTH = 150
+MAX_CONTEXT_SEGMENTS = 3
+MAX_GEMMA_LENGTH = 128
 
 # --- Global Conversation History ---
 conversation_history = []
@@ -65,7 +65,7 @@ def generate_response(text: str, model_gemma, tokenizer_gemma, device) -> str: #
     try:
         # Gemma 3 chat template format
         messages = [{"role": "user", "content": text}]
-        input = tokenizer_gemma.apply_chat_template(messages, return_tensors="pt").to(device)
+        input = tokenizer_gemma.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(device)
         generation_config = GenerationConfig(
             max_new_tokens=MAX_GEMMA_LENGTH,
             early_stopping=True,