Update app.py

app.py

@@ -31,8 +31,8 @@ def chat(user_input):
         output = model.generate(
             **input_ids,
             max_new_tokens=256,
-            do_sample=False,
-            temperature=0.0,
+            do_sample=True,
+            temperature=0.7,
             repetition_penalty=1.2,
             no_repeat_ngram_size=5,
             top_k=50,