Update app.py

app.py

@@ -37,7 +37,8 @@ def generate(
     message: str,
     chat_history: list[dict],
     max_new_tokens: int = 1024,
-    temperature: float = 0.0,
+    do_sample: bool = False,
+    temperature: float = 0.6,
     top_p: float = 0.9,
     top_k: int = 50,
     repetition_penalty: float = 1.2,
@@ -56,7 +57,7 @@ def generate(
         {"input_ids": input_ids},
         streamer=streamer,
         max_new_tokens=max_new_tokens,
-        do_sample=True,
+        do_sample=do_sample,
         top_p=top_p,
         top_k=top_k,
         temperature=temperature,
@@ -82,6 +83,7 @@ chat_interface = gr.ChatInterface(
             step=1,
             value=DEFAULT_MAX_NEW_TOKENS,
         ),
+        gr.Checkbox(label="Do Sample"),
         gr.Slider(
             label="Temperature",
             minimum=0.0,