Update app.py

app.py

@@ -42,11 +42,11 @@ def run_generation(user_text, top_p, temperature, top_k, max_new_tokens):
     generate_kwargs = dict(
         inputs,
         streamer=streamer,
-        max_new_tokens=max_new_tokens,
+        max_new_tokens=250,
         do_sample=True,
         repetition_penalty=1.5,
-        temperature=float(temperature),
-        top_k=top_k,
+        temperature=0.7,
+        top_k=0.2,
     )
     t = Thread(target=model.generate, kwargs=generate_kwargs)
     t.start()
@@ -101,12 +101,12 @@ with gr.Blocks() as demo:
 
     user_text.submit(
         run_generation,
-        [user_text,temperature,top_k, max_new_tokens],
+        [user_text],
          model_output,
     )
     button_submit.click(
         run_generation,
-        [user_text, temperature, top_k, max_new_tokens],
+        [user_text],
         model_output,
     )