Updates

app.py

@@ -45,7 +45,6 @@ class ChatBot:
         self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
         self.tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat-v1.0", use_fast=False)
         self.model = AutoModelForCausalLM.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat-v1.0")
-        self.model = torch.quantization.quantize_dynamic(self.model, {torch.nn.Linear}, dtype=torch.qint8)
 
         # Parler-TTS Setup
         self.tts_model = ParlerTTSForConditionalGeneration.from_pretrained("doublesizebed/parler-tts-mini-malay")
@@ -65,7 +64,7 @@ class ChatBot:
                 "Q: What is photosynthesis?\n"
                 "Answer: Photosynthesis lets plants convert sunlight into energy. Which plants interest you most?\n\n"
                 "Q: How do I make tea?\n"
-                "Answer: Steep tea leaves in hot water for 3–5 minutes, then serve. Do you prefer green or black tea?\n\n"
+                "Answer: Steep tea leaves in hot water for 3-5 minutes, then serve. Do you prefer green or black tea?\n\n"
             )
             # 3) The actual user query
             query = f"Q: {user_question}\nAnswer:"