Update app.py

app.py

@@ -66,7 +66,7 @@ def llm_run(prompt, max_length, top_p, temprature, top_k, messages):
     )
     generate_kwargs = dict(
         max_length=int(max_length),top_p=float(top_p), do_sample=True,
-        top_k=int(top_k), streamer=streamer, temperature=int(temprature), repetition_penalty=1.2
+        top_k=int(top_k), streamer=streamer, temperature=float(temprature), repetition_penalty=1.2
     )
 
     t = Thread(target=model.generate,  args=(input_ids,), kwargs=generate_kwargs)