Update app.py

app.py

@@ -12,7 +12,7 @@ def format_prompt(message, history):
   return prompt
 
 def generate(
-    prompt, history=[], temperature=0.2, max_new_tokens=35000, top_p=0.95, repetition_penalty=1.0,
+    prompt, history=[], temperature=0.2, max_new_tokens=2000, top_p=0.95, repetition_penalty=1.0,
 ):
     temperature = float(temperature)
     if temperature < 1e-2:
@@ -46,7 +46,8 @@ def health():
 def search_route():
     data = request.get_json()
     prompt = data.get('prompt', '')
-    return Response(generate(prompt), content_type='text/plain', status=200, direct_passthrough=True)
+    truncated_prompt = original_prompt[:32768]
+    return Response(generate(truncated_prompt), content_type='text/plain', status=200, direct_passthrough=True)
 
 if __name__ == '__main__':
     app.run(debug=False, host='0.0.0.0', port=7860)