Update app.py

app.py

@@ -60,14 +60,16 @@ class Llama3Demo:
 
     
 ##################################################################
-    def generate_response(self, prompt: str, max_new_tokens: int = 512) -> str:
-        formatted_prompt = f"""<|system|>You are a helpful AI assistant.</s>
-<|user|>{prompt}</s>
-<|assistant|>"""
+    def generate_response(self, prompt: str, max_new_tokens: int = 512, temperature: float = 0.6, 
+                         top_p: float = 0.85, repetition_penalty: float = 1.2, top_k: int = 50) -> str:
+        formatted_prompt = f"""<|system|>You are a helpful AI assistant. Always provide accurate, 
+    detailed, and well-reasoned responses. If you're unsure about something, acknowledge the uncertainty. 
+    Break down complex topics into clear explanations.</s>
+    <|user|>{prompt}</s>
+    <|assistant|>"""
         
         inputs = self.tokenizer(formatted_prompt, return_tensors="pt").to(self.model.device)
         
-        # Asegurar que tenemos un pad_token_id válido
         if self.tokenizer.pad_token_id is None:
             self.tokenizer.pad_token_id = self.tokenizer.eos_token_id
         
@@ -76,10 +78,12 @@ class Llama3Demo:
                 **inputs,
                 max_new_tokens=max_new_tokens,
                 num_return_sequences=1,
-                temperature=0.7,
+                temperature=temperature,
                 do_sample=True,
-                top_p=0.9,
-                pad_token_id=self.tokenizer.pad_token_id  # Explícitamente establecer pad_token_id
+                top_p=top_p,
+                top_k=top_k,
+                repetition_penalty=repetition_penalty,
+                pad_token_id=self.tokenizer.pad_token_id
             )
             
             torch.cuda.empty_cache()