Update app.py

app.py

@@ -4,11 +4,6 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
 
 tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-mamba-7b-instruct")
 model = AutoModelForCausalLM.from_pretrained("tiiuae/falcon-mamba-7b-instruct")
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-
-
 
 def respond(
     message,
@@ -32,7 +27,7 @@ def respond(
 
     for message in client.chat_completion(
         messages,
-        max_tokens=max_tokens,
+        max_tokens=1024,
         stream=True,
         temperature=temperature,
         top_p=top_p,