lowered model temperature to see if CoT reasoning in controllable

app.py

@@ -85,7 +85,7 @@ class RoleAgent:
             **inputs,
             max_new_tokens=128,
             do_sample=True,
-            temperature=0.7,
+            temperature=0.3,
             pad_token_id=self.tokenizer.eos_token_id,
         )
         response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)