File size: 193,998 Bytes
6fa4bc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
{
    "paper_id": "2022",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T01:11:39.935405Z"
    },
    "title": "Foundation Models of Scientific Knowledge for Chemistry: Opportunities, Challenges and Lessons Learned",
    "authors": [
        {
            "first": "Sameera",
            "middle": [],
            "last": "Horawalavithana",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Pacific Northwest National Laboratory",
                "location": {
                    "settlement": "Richland",
                    "region": "WA"
                }
            },
            "email": ""
        },
        {
            "first": "Ellyn",
            "middle": [],
            "last": "Ayton",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Pacific Northwest National Laboratory",
                "location": {
                    "settlement": "Richland",
                    "region": "WA"
                }
            },
            "email": ""
        },
        {
            "first": "Shivam",
            "middle": [],
            "last": "Sharma",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Pacific Northwest National Laboratory",
                "location": {
                    "settlement": "Richland",
                    "region": "WA"
                }
            },
            "email": ""
        },
        {
            "first": "Scott",
            "middle": [],
            "last": "Howland",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Pacific Northwest National Laboratory",
                "location": {
                    "settlement": "Richland",
                    "region": "WA"
                }
            },
            "email": ""
        },
        {
            "first": "Megha",
            "middle": [],
            "last": "Subramanian",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Pacific Northwest National Laboratory",
                "location": {
                    "settlement": "Richland",
                    "region": "WA"
                }
            },
            "email": ""
        },
        {
            "first": "Scott",
            "middle": [],
            "last": "Vasquez",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Pacific Northwest National Laboratory",
                "location": {
                    "settlement": "Richland",
                    "region": "WA"
                }
            },
            "email": ""
        },
        {
            "first": "Robin",
            "middle": [],
            "last": "Cosbey",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Pacific Northwest National Laboratory",
                "location": {
                    "settlement": "Richland",
                    "region": "WA"
                }
            },
            "email": ""
        },
        {
            "first": "Maria",
            "middle": [],
            "last": "Glenski",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Pacific Northwest National Laboratory",
                "location": {
                    "settlement": "Richland",
                    "region": "WA"
                }
            },
            "email": ""
        },
        {
            "first": "Svitlana",
            "middle": [],
            "last": "Volkova",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Pacific Northwest National Laboratory",
                "location": {
                    "settlement": "Richland",
                    "region": "WA"
                }
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Foundation models pre-trained on large corpora demonstrate significant gains across many natural language processing tasks and domains e.g., law, healthcare, education, etc. However, only limited efforts have investigated the opportunities and limitations of applying these powerful models to science and security applications. In this work, we develop foundation models of scientific knowledge for chemistry to augment scientists with the advanced ability to perceive and reason at scale previously unimagined. Specifically, we build large-scale (1.47B parameter) general-purpose models for chemistry that can be effectively used to perform a wide range of in-domain and out-of-domain tasks. Evaluating these models in a zero-shot setting, we analyze the effect of model and data scaling, knowledge depth, and temporality on model performance in context of model training efficiency. Our novel findings demonstrate that (1) model size significantly contributes to the task performance when evaluated in a zero-shot setting; (2) data quality (aka diversity) affects model performance more than data quantity; (3) similarly, unlike previous work (Luu et al., 2021) temporal order of the documents in the corpus boosts model performance only for specific tasks, e.g., SciQ; and (4) models pre-trained from scratch perform better on in-domain tasks than those tuned from general-purpose models like Open AI's GPT-2.",
    "pdf_parse": {
        "paper_id": "2022",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Foundation models pre-trained on large corpora demonstrate significant gains across many natural language processing tasks and domains e.g., law, healthcare, education, etc. However, only limited efforts have investigated the opportunities and limitations of applying these powerful models to science and security applications. In this work, we develop foundation models of scientific knowledge for chemistry to augment scientists with the advanced ability to perceive and reason at scale previously unimagined. Specifically, we build large-scale (1.47B parameter) general-purpose models for chemistry that can be effectively used to perform a wide range of in-domain and out-of-domain tasks. Evaluating these models in a zero-shot setting, we analyze the effect of model and data scaling, knowledge depth, and temporality on model performance in context of model training efficiency. Our novel findings demonstrate that (1) model size significantly contributes to the task performance when evaluated in a zero-shot setting; (2) data quality (aka diversity) affects model performance more than data quantity; (3) similarly, unlike previous work (Luu et al., 2021) temporal order of the documents in the corpus boosts model performance only for specific tasks, e.g., SciQ; and (4) models pre-trained from scratch perform better on in-domain tasks than those tuned from general-purpose models like Open AI's GPT-2.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "The emergence of foundation models (Bommasani et al., 2021) such as large-scale autoencoding models (e.g., BERT (Devlin et al., 2018) , RoBERTa ) and autoregressive language models (e.g., GPT-2 (Radford et al., 2019) , GPT-3 (Brown et al., 2020) , Megatron-Turing (Smith et al., 2022) and Gopher (Rae et al., 2021) ) as well as multimodal vision and language models, such as FLAVA (Singh et al., 2021) and Perceiver (Jaegle et al., 2021) , established a paradigm shift in Artificial Intelligence (AI). These foundation models, also called neural platforms, are built using self-supervised pretraining at scale. They are then able to be easily adapted to a wide range of downstream tasks via transfer learning (Bommasani et al., 2021) and finetuning .",
                "cite_spans": [
                    {
                        "start": 35,
                        "end": 59,
                        "text": "(Bommasani et al., 2021)",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 112,
                        "end": 133,
                        "text": "(Devlin et al., 2018)",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 188,
                        "end": 216,
                        "text": "GPT-2 (Radford et al., 2019)",
                        "ref_id": null
                    },
                    {
                        "start": 219,
                        "end": 245,
                        "text": "GPT-3 (Brown et al., 2020)",
                        "ref_id": null
                    },
                    {
                        "start": 264,
                        "end": 284,
                        "text": "(Smith et al., 2022)",
                        "ref_id": "BIBREF52"
                    },
                    {
                        "start": 289,
                        "end": 314,
                        "text": "Gopher (Rae et al., 2021)",
                        "ref_id": null
                    },
                    {
                        "start": 381,
                        "end": 401,
                        "text": "(Singh et al., 2021)",
                        "ref_id": "BIBREF51"
                    },
                    {
                        "start": 416,
                        "end": 437,
                        "text": "(Jaegle et al., 2021)",
                        "ref_id": "BIBREF21"
                    },
                    {
                        "start": 709,
                        "end": 733,
                        "text": "(Bommasani et al., 2021)",
                        "ref_id": "BIBREF7"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "The wide community adoption of foundation models can be explained by their key properties, two of which are emergent behavior and homogenization -which also make foundation models appealing for adaption across science and security domains. Emergence, or emergent behavior, reflect new behaviors that a model introduces or is capable of that it was not explicitly trained to perform. Homogenization is the consolidation of methods for building machine learning systems across a wide range of tasks. Another key advantage of scaling language models is that they perform competitively on language tasks using in-context learning without fine-tuning or gradient updates. Thus, in-context learning allows foundation models to be effectively used across new downstream tasks with only simple instructions and a few optional examples.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In this work we focus on a science domain (chemistry) and demonstrate the value and limitations of large-scale language models evaluated across a wide range of in-domain (science-focused) and out-of-domain tasks. Unlike the majority of work on foundation models that focuses on pretraining these models on book corpora, web pages, Wikipedia and mixed sources, e.g., the Pile , we pretrain our models on scientific literature. Using scientific literature presents unique opportunities and challenges. Opportunities include the scale and diversity of scientific literature, the explicit structure, and explicit alignment across different modalities in the papers, e.g., table and figure references. Challenges include limited benchmarks that can be used to perform model evaluation, model prompting and interactions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "There are three major contributions of this work: (1) we collect and release a 0.67TB dataset covering research publication data across 10+ sources for chemistry; (2) we release 28 auto-regressive foundation models for chemistry that have been pretrained from scratch; and (3) we present a rigorous evaluation of model performance on 15+ indomain and out-of-domain tasks that investigates the effects of model and data scaling, knowledge depth (aka diversity), and temporal order on performance as described in research questions below.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "(RQ1) Science-Focused Benchmarks What are the strengths and weaknesses of foundation models pretrained on scientific literature when evaluated on out-of-domain vs. in-domain tasks?",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "(RQ2) Scaling Effect How does model scale affect the downstream performance? Do neural scaling laws presented in (Kaplan et al., 2020) hold for the foundation models for science?",
                "cite_spans": [
                    {
                        "start": 113,
                        "end": 134,
                        "text": "(Kaplan et al., 2020)",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "(RQ3) Diversity Effect How does the depth of scientific knowledge, e.g., from paper abstracts vs. full text, affect downstream performance? (RQ4) Temporal Effect How does the recency of scientific knowledge, e.g., when manipulating the temporal order of the documents processed by the model, affect downstream performance?",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In this section we summarize previous efforts in two categories: mixed-domain continual pretraining that continues pretraining of a base model on domain data and in-domain pretraining from scratch that pretrains a from scratch on domain data. We present a model summary in Table 1 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 273,
                        "end": 280,
                        "text": "Table 1",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Many efforts have focused on continual pretraining of a BERT (Devlin et al., 2018) base model. Several models have been developed for the biomedical domain and the most frequently used corpora for domain-specific continual preraining are PubMed abstracts and PubMed Central full-text articles (PMC) (Lee et al., 2020; Peng et al., 2019; Phan et al., 2021) . In the Chemistry domain, Guo et al. (2021) performed continual pretraining of a base BERT model on 200K chemistry journal articles for product extraction (ChemBERT) and reaction role labeling (ChemRxnBERT).",
                "cite_spans": [
                    {
                        "start": 61,
                        "end": 82,
                        "text": "(Devlin et al., 2018)",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 299,
                        "end": 317,
                        "text": "(Lee et al., 2020;",
                        "ref_id": "BIBREF26"
                    },
                    {
                        "start": 318,
                        "end": 336,
                        "text": "Peng et al., 2019;",
                        "ref_id": "BIBREF40"
                    },
                    {
                        "start": 337,
                        "end": 355,
                        "text": "Phan et al., 2021)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Mixed-Domain Continual Pretraining",
                "sec_num": null
            },
            {
                "text": "In-Domain Pretraining from Scratch Previous work has shown that pretraining models from scratch on domain-specific data has a significant benefit over continual pretraining of generaldomain language models (Gu et al., 2021) . This is mainly due to the availability of in-domain data for both generating the vocabulary and pretraining. SciBERT (Beltagy et al., 2019) is pretrained according to this procedure using the vocabulary generated from computer science and biomedical domains. PubMedBERT (Gu et al., 2021) is another example of pretraining the base BERT model from scratch using PubMed. Unlike any previous work, we use both continual and from scratch pretraining to build the largest foundation model for Chemistry (1.47B) on the largest (0.67TB) and the most diverse corpus (10+ sources) collected to date.",
                "cite_spans": [
                    {
                        "start": 206,
                        "end": 223,
                        "text": "(Gu et al., 2021)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 343,
                        "end": 365,
                        "text": "(Beltagy et al., 2019)",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 496,
                        "end": 513,
                        "text": "(Gu et al., 2021)",
                        "ref_id": "BIBREF18"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Mixed-Domain Continual Pretraining",
                "sec_num": null
            },
            {
                "text": "Unlike the majority of related models that rely on a base BERT (or variant) model, we adapt the Open-AI's GPT-2 transformer decoder architecture (Radford et al., 2019) to train autoregressive language models for Chemistry. To understand the impact of model size (RQ2), we experiment with four different Transformer sizes: small (S), medium (M), large (L), and extra-large (XL). These models differ in the number of decoder layers, hidden size of the model, and the number of attention heads in transformer blocks as shown in Table 2 .",
                "cite_spans": [
                    {
                        "start": 145,
                        "end": 167,
                        "text": "(Radford et al., 2019)",
                        "ref_id": "BIBREF44"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 525,
                        "end": 532,
                        "text": "Table 2",
                        "ref_id": "TABREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Model Pretraining",
                "sec_num": "3"
            },
            {
                "text": "Our experiments leverage the GPT-NeoX Python library (Andonian et al., 2021) developed with Megatron (Shoeybi et al., 2019) and Deep-Speed (Rasley et al., 2020) . We optimize the autoregressive log-likelihood (i.e., cross-entropy loss) averaged over a 2048-token context. We set the micro batch size per GPU as 4, and the learning rate to 2 \u00d7 10 \u22124 , and rely on the cosine decay. We use an Adam optimizer with \u03b2 1 = 0.9, \u03b2 2 = 0.99, and \u03c3 = 10 \u22128 and clip the gradient norm at 1.0. In addition, ZeRO optimizer (Rajbhandari et al., 2019) was used to reduce memory footprint by distributing optimizer states across several processes.",
                "cite_spans": [
                    {
                        "start": 53,
                        "end": 76,
                        "text": "(Andonian et al., 2021)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 101,
                        "end": 123,
                        "text": "(Shoeybi et al., 2019)",
                        "ref_id": "BIBREF50"
                    },
                    {
                        "start": 139,
                        "end": 160,
                        "text": "(Rasley et al., 2020)",
                        "ref_id": "BIBREF48"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Model Pretraining",
                "sec_num": "3"
            },
            {
                "text": "To reduce memory and increase training throughput, we use mixed-precision training (Rasley et al., 2020) and the parallel attention and feed-forward implementations available in GPT-NeoX (Black et al., 2022) . We also use the Rotary positional embeddings instead of the learned positional embeddings used in the GPT-2 model (Radford et al., 2019) because they offer performance advantages in tasks with longer texts by capturing relative position dependency in self-attention. Our models are pretrained across multiple workers with data parallelism. As the largest model in our experiments fit on a single GPU, we didn't use the model (tensor) or pipeline parallelism. Models are pretrained from scratch for a total of 320K steps. The original GPT-2 models are fine-tuned for 150K steps. We perform experiments in a single DGX-A100 machine with 8 80Gb GPUs.",
                "cite_spans": [
                    {
                        "start": 83,
                        "end": 104,
                        "text": "(Rasley et al., 2020)",
                        "ref_id": "BIBREF48"
                    },
                    {
                        "start": 187,
                        "end": 207,
                        "text": "(Black et al., 2022)",
                        "ref_id": "BIBREF6"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Model Pretraining",
                "sec_num": "3"
            },
            {
                "text": "We collected a large corpus of 53.45 million chemistry-focused scientific articles and abstracts, resulting in 670GB of text data. As shown in Table 3, our corpus was collected from 10 different data sources: Arxiv, Aminer (AMiner), CORD-19 (Wang et al., 2020b) , CORE (Pontika et al., 2016) , Microsoft Academic Graph (MAG) (Wang et al., 2020a) , OSTI, PubMed (abstracts and fulltexts), and the Web of Science (WoS). See Appendix A for full data descriptions. Because the data sources we relied on comprise research publications from many science domains, we sampled articles using a list of domain-specific keywords for chemistry to create the dataset summarized in Table 3 . These keywords were extracted by using a Correlation Explanation (Gallagher et al., 2017) topic model followed by manual filtering by subject matter experts. This resulted in a list of more than 1K chemistry-related entities, ranging from compound names like ethyl acetate, methyl methacrylate, sulfoxide, etc. to experiment and procedures like tunneling microscopy, neutralization, enzymatic hydrolysis, etc. Data Cleaning Recent research has shown that duplicates in training data can significantly impact the downstream task performance of LLMs (Lee et al., 2021; Carlini et al., 2022) . To this end, we performed deduplication of our corpus based on overlap of titles within and across data sources. We processed titles to strip punctuation and casefold and considered two articles A 1 and A 2 to be duplicates if they had the same processed title. With this technique, we were able to remove significant amounts of duplicate scientific articles both within and across sources. The deduplication process reduced our corpus from 875GB to 670GB (67.8M to 53.5M publications), removing 14.3M duplicates.",
                "cite_spans": [
                    {
                        "start": 233,
                        "end": 261,
                        "text": "CORD-19 (Wang et al., 2020b)",
                        "ref_id": null
                    },
                    {
                        "start": 269,
                        "end": 291,
                        "text": "(Pontika et al., 2016)",
                        "ref_id": "BIBREF43"
                    },
                    {
                        "start": 325,
                        "end": 345,
                        "text": "(Wang et al., 2020a)",
                        "ref_id": "BIBREF54"
                    },
                    {
                        "start": 1226,
                        "end": 1244,
                        "text": "(Lee et al., 2021;",
                        "ref_id": null
                    },
                    {
                        "start": 1245,
                        "end": 1266,
                        "text": "Carlini et al., 2022)",
                        "ref_id": "BIBREF9"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 668,
                        "end": 675,
                        "text": "Table 3",
                        "ref_id": "TABREF2"
                    }
                ],
                "eq_spans": [],
                "section": "Data Collection and Processing",
                "sec_num": "4"
            },
            {
                "text": "Tokenization As used in GPT-2 model, we use a Byte Pair Encoding (BPE) tokenizer. We train BPE tokenizers for each data sample with a vocabulary size of 64K as preliminary experiments varying vocabulary sizes from 64K to 256K for smaller scale model pretraining did not show significant differences in performance. We compare the GPT-2 vocabulary generated from the WebText and the in-domain vocabularies generated from our corpora and find that the in-domain vocabulary breaks chemical entities into fewer tokens. For example, dimethylnitroxide was tokenized into #dimethyl, #nitr, #oxide using the in-domain vocabulary and #dim, #ethyl, #nit, #rox, #ide using the GPT-2 vocabulary.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Data Collection and Processing",
                "sec_num": "4"
            },
            {
                "text": "This section presents the analysis of 28 pretrained models evaluated on 15+ in-domain and out-ofdomain downstream tasks (RQ1, Section 5.1). We investigate the effects of model and data scaling (RQ2, Section 5.2), knowledge diversity (RQ3, Section 5.3), and temporal order (RQ4, Section 5.4) on the downstream performance. We also compare the results from continual vs. from scratch pretraining (Section 5.5) and present the analysis of large-scale training efficiency (Section 5.6).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Analysis and Results",
                "sec_num": "5"
            },
            {
                "text": "Baseline Models As we use a similar model architecture, we identify Open AI's GPT-2 (Radford et al., 2019) as a baseline comparison model. We compare our performance with four variants of the original GPT-2 models, corresponding to small (S), medium (M), large (L), and extra-large (XL) sized transformer architectures shown in Table 2 . We note that GPT-2 models were pretrained on Web-Text -8 million web documents (40Gb). Thus, we also include a base GPT-2 model (medium) that has been updated with continual pretraining using our Combined \u2329A+FT\u232a dataset.",
                "cite_spans": [
                    {
                        "start": 84,
                        "end": 106,
                        "text": "(Radford et al., 2019)",
                        "ref_id": "BIBREF44"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 328,
                        "end": 335,
                        "text": "Table 2",
                        "ref_id": "TABREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Analysis and Results",
                "sec_num": "5"
            },
            {
                "text": "Our Models We pretrained models with individual datasets (AMiner, CORE, MAG, PubMed, S2ORC, WOS) and combined abstracts and fulltexts. Our goal is to systematically study data biases in the model performance when pretraining models with individual datasets. For example, PubMed publications cover mostly bio-medicinal terms (Gu et al., 2021) , while the majority of S2ORC publications are from medicine, biology, physics, and mathematics (Lo et al., 2020) . We only use 4 GPUs for the models pretrained with individual datasets and 8 GPUs for the rest. This is to control the number of tokens seen during model pretraining (320,000 steps * 4 GPUs * 4 micro batch size * 2,048 context size = 10B tokens) relative to the maximum number of tokens available in the respective datasets (as reported in Table 3 ). We also trained one XL (4x) model with 4x larger batch size than what used in XL model to evaluate the impact of the number of training tokens.",
                "cite_spans": [
                    {
                        "start": 324,
                        "end": 341,
                        "text": "(Gu et al., 2021)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 438,
                        "end": 455,
                        "text": "(Lo et al., 2020)",
                        "ref_id": "BIBREF33"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 797,
                        "end": 804,
                        "text": "Table 3",
                        "ref_id": "TABREF2"
                    }
                ],
                "eq_spans": [],
                "section": "Analysis and Results",
                "sec_num": "5"
            },
            {
                "text": "We evaluate our models using several benchmarks to assess the effectiveness in both in-domain and out-of-domain tasks. The benchmarks we include are described in Appendix B. We use the lmevaluation-harness Python repository (Gao et al., Table 4 : Downstream Zero-shot In-Domain Task Performance. We use \u2021 to indicate the baseline model tuned from the base GPT-2 model. Pile performance is reported using perplexity, with all other tasks reported using accuracy. We highlight the top-4 performance per task in bold, with top performance indicated with an underline. XL (4x) model is trained with 4x larger batch size that used in other models. 2021) for the benchmark implementation.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 237,
                        "end": 244,
                        "text": "Table 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Zero-shot Performance",
                "sec_num": "5.1"
            },
            {
                "text": "In-domain Evaluation We consider five existing chemistry benchmarks, specifically Hendryck-sTest (Hendrycks et al., 2020) for high school (HT-HC) and college (HT-CC) levels, and sciencefocused -ARC , SciQ (Welbl et al., 2017) , OpenBookQA (Mihaylov et al., 2018) , Pile-PubMed-Abstracts ). As shown in Table 4 , one or more of our models outperform baseline GPT-2 models for the two chemistry tasks, general science QA (SciQ) and the sciencefocused language modelling. Of the remaining tasks, our models perform within 1-4% of GPT-2 baselines.",
                "cite_spans": [
                    {
                        "start": 97,
                        "end": 121,
                        "text": "(Hendrycks et al., 2020)",
                        "ref_id": null
                    },
                    {
                        "start": 205,
                        "end": 225,
                        "text": "(Welbl et al., 2017)",
                        "ref_id": "BIBREF56"
                    },
                    {
                        "start": 239,
                        "end": 262,
                        "text": "(Mihaylov et al., 2018)",
                        "ref_id": "BIBREF36"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 302,
                        "end": 309,
                        "text": "Table 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Zero-shot Performance",
                "sec_num": "5.1"
            },
            {
                "text": "Out-of-domain Evaluation We evaluate outof-domain performance using 9 commonly used LLM benchmarks: BoolQ (Clark et al., 2019) , CB (De Marneffe et al., 2019) , WIC (Pilehvar and Camacho-Collados, 2018), WSC (Levesque et al., 2012 ), MathQA (Amini et al., 2019) , PIQA (Bisk et al., 2020) , PubMedQA , Lambada (Paperno et al., 2016) and WikiText (Merity et al., 2016) . As shown in Table 5 , our models outperform baseline GPT-2 models for CB, WIC and WSC and match the best accuracy for BoolQ but the GPT-2 baselines outperform on the remaining tasks, particularly Lambada and Wikitext -the two general language modeling tasks.",
                "cite_spans": [
                    {
                        "start": 106,
                        "end": 126,
                        "text": "(Clark et al., 2019)",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 136,
                        "end": 158,
                        "text": "Marneffe et al., 2019)",
                        "ref_id": "BIBREF13"
                    },
                    {
                        "start": 208,
                        "end": 230,
                        "text": "(Levesque et al., 2012",
                        "ref_id": "BIBREF28"
                    },
                    {
                        "start": 241,
                        "end": 261,
                        "text": "(Amini et al., 2019)",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 269,
                        "end": 288,
                        "text": "(Bisk et al., 2020)",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 310,
                        "end": 332,
                        "text": "(Paperno et al., 2016)",
                        "ref_id": "BIBREF39"
                    },
                    {
                        "start": 346,
                        "end": 367,
                        "text": "(Merity et al., 2016)",
                        "ref_id": "BIBREF35"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 382,
                        "end": 389,
                        "text": "Table 5",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Zero-shot Performance",
                "sec_num": "5.1"
            },
            {
                "text": "Previous work (Kaplan et al., 2020) has shown that upstream cross entropy loss scales as a power-law with model size, dataset size, and the amount of compute. In this section, we revisit these claims on scaling Transformer architectures.",
                "cite_spans": [
                    {
                        "start": 14,
                        "end": 35,
                        "text": "(Kaplan et al., 2020)",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Scaling Effect",
                "sec_num": "5.2"
            },
            {
                "text": "Analyzing upstream cross entropy loss During pretraining, we group each dataset into training/validation/test (949/50/1) splits. We report the model performance on validation data using cross entropy loss in nats. This measure will be averaged over the 2048-token context. We find that the cross entropy loss decreases as we increase the model size (as shown in Figure 2 ). Larger models reach a given loss value in a higher rate than the smaller models. This observation illustrates the relationship between model performance (as measured by the upstream cross entropy loss) and model size, confirming (Kaplan et al., 2020) .",
                "cite_spans": [
                    {
                        "start": 603,
                        "end": 624,
                        "text": "(Kaplan et al., 2020)",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 362,
                        "end": 370,
                        "text": "Figure 2",
                        "ref_id": "FIGREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Scaling Effect",
                "sec_num": "5.2"
            },
            {
                "text": "Analyzing downstream task performance Can we speculate downstream task performance of a model from the pretraining performance? First, we find that the models perform considerably well on Pile in comparison to the Lambada or WikiText.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Scaling Effect",
                "sec_num": "5.2"
            },
            {
                "text": "There is a 48% performance advantage in this task over the best performing baseline GPT-2 model. This may be due to the models capturing scientific language better than general language. It is im- portant to note that we exclude PubMed Abstracts in the individual data collection to avoid potential contamination between the training and Pile testing data. As shown in Table 4 , larger models perform well on these language modeling tasks. Second, we noticed that the XL (4x) model trained for more tokens performs significantly better than the similar sized XL model. Specifically, XL (4x) model was trained with 128 total batch size compared to the 32 total batch size used in XL model. XL (4x) model achieves the lowest Lambada and WikiText perplexity values across all our models trained from scratch (as shown in Table 5 ). The same model also achieves the best SciQ performance with 0.84 accuracy and comparable in other tasks performance with the XL model. This experiment highlights the importance of training models with larger batch size. We note that the baseline models (Radford et al., 2019) were trained with 4x larger batch size (total batch size 512) than what used in XL (4x) model. We believe that the XL (4x) model can reach the similar perplexity values when trained for this data scale.",
                "cite_spans": [
                    {
                        "start": 1082,
                        "end": 1104,
                        "text": "(Radford et al., 2019)",
                        "ref_id": "BIBREF44"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 369,
                        "end": 376,
                        "text": "Table 4",
                        "ref_id": null
                    },
                    {
                        "start": 818,
                        "end": 825,
                        "text": "Table 5",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Scaling Effect",
                "sec_num": "5.2"
            },
            {
                "text": "Third, we find that zero-shot task performance in SciQ, HT-CC and ARC-E increases as we increase the model size (see Table 5 ). However, there is no clear relationship between the task performance and the model sizes in the rest of benchmark datasets. We suggest that pretraining performance may not be the ideal indicator to speculate the overall downstream task performance, especially in the zero-shot setting. However, model size significantly contributes to the task performance.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 117,
                        "end": 124,
                        "text": "Table 5",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Scaling Effect",
                "sec_num": "5.2"
            },
            {
                "text": "While abstracts often provide a summary of scientific publications, the full text contains more details.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Diversity Effect",
                "sec_num": "5.3"
            },
            {
                "text": "In this section, we analyze the performance of models trained on paper abstracts versus full texts.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Diversity Effect",
                "sec_num": "5.3"
            },
            {
                "text": "First, the XL models trained with the combined abstract dataset achieve the lowest perplexity score (22.77) on the Pile -a 45% performance advantage over the full text version. There are might be several factors that contribute to this, but one may be the focused language in abstracts. Second, the model trained with the combined abstracts achieves the second best accuracy (0.83 in comparison to 0.79 for the full text model) in SciQ. Some of the models pretrained on individual abstract data achieve comparable performance in SciQ, e.g., MAG and AMiner models achieve 0.8 and 0.78 accuracy, respectively. We believe the diversity of scientific knowledge provided from the abstract data is useful since SciQ questions span biology, chemistry, earth science, and physics.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Diversity Effect",
                "sec_num": "5.3"
            },
            {
                "text": "Third, we compare model performance trained with abstracts vs. full texts in the HT task and see that the best accuracy is achieved using the MAG and S2ORC datasets rather than the combined abstracts. This suggests the importance of contextual knowledge provided by different data sources.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Diversity Effect",
                "sec_num": "5.3"
            },
            {
                "text": "Finally, combined full text model performs better than the model trained with the abstracts in all outof-domain tasks except PIQA. This performance difference may be due to the more expressive and diverse language presented in the full texts than in the abstracts. Thus, expanding full text coverage may improve out-of-domain task generalization.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Diversity Effect",
                "sec_num": "5.3"
            },
            {
                "text": "Scientific knowledge evolves over time reflecting new research ideas, innovations, and findings. In this section, we test how continual pretraining on temporal-aligned scientific publications impacts downstream performance. For this experiment, we maintain two variants of the MAG dataset with random-ordered and temporal-ordered articles, splitting each into ten equal subsets. We continue pretraining a base medium (M) sized model iteratively with the subsets in the order they appeared in the respective data variant. For example, in the temporally-aligned experiments, we first pretrain a model with 3.4M (10%) articles from before 1978, and then use it as the base model to continue pretraining with another 3.4M (10%) articles from between 1978 and 1989. We train the initial model for 150K steps and each subsequent model for 10K steps with additional data. Figure 3 shows the performance of model checkpoints across in-domain and out-of-domain tasks.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 865,
                        "end": 873,
                        "text": "Figure 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Temporal Effect",
                "sec_num": "5.4"
            },
            {
                "text": "There are two key findings. First, SciQ and ARC-E zero-shot task performances improve over time with the models trained with temporally-ordered scientific texts (as shown in Figure 3b ). For example, SciQ accuracy improves from 0.64 to 0.73 from the base model checkpoint to the final model checkpoint. Similarly, ARC-E accuracy improves from 0.43 to 0.45. This is due to the temporal order of the knowledge acquired by the model. When the model was pretrained with random-ordered data subsets, we observe only a slight (< 1%) performance increase (as shown in Figure 3a) .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 174,
                        "end": 183,
                        "text": "Figure 3b",
                        "ref_id": null
                    },
                    {
                        "start": 561,
                        "end": 571,
                        "text": "Figure 3a)",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Temporal Effect",
                "sec_num": "5.4"
            },
            {
                "text": "There are mixed patterns in performance across out-of-domain tasks. For example, a slight performance increase in the PIQA, CB, PubMedQA, and WIC over time with the models trained with temporally-ordered scientific texts. On the other hand, there is a performance drop in the BoolQ and WSC over time. This may be due to the catastrophic forgetting prevalent in continual learning (Ramasesh et al., 2021) . Future work will in- Figure 3 : The effect of temporal order of publications during pretraining. We align publications in the MAG corpus by year and split them into ten equal subsets. We repeat the process in a randomly-ordered corpus for comparison, recording model checkpoints after performing continual pretraining on each data subset.",
                "cite_spans": [
                    {
                        "start": 380,
                        "end": 403,
                        "text": "(Ramasesh et al., 2021)",
                        "ref_id": "BIBREF47"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 427,
                        "end": 435,
                        "text": "Figure 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Temporal Effect",
                "sec_num": "5.4"
            },
            {
                "text": "vestigate other confounding factors that may contribute to this performance patterns.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Effect",
                "sec_num": "5.4"
            },
            {
                "text": "In this section, we test whether the continual pretraining of a base GPT model with additional domain-specific data is helpful in the downstream task performance. We report the zero-shot performance of the tuned model across in-domain (Table 4) and out-of-domain (Table 5 ) tasks. We have two main observations from this experiment.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 263,
                        "end": 271,
                        "text": "(Table 5",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Continual vs. From Scratch Pretraining",
                "sec_num": "5.5"
            },
            {
                "text": "First, fine-tuned models fall behind other baselines in a majority of in-domain tasks. HT-CC is the only in-domain task that the tuned model outperforms the rest of models, yet fails to outperform the best performing model trained from scratch.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Continual vs. From Scratch Pretraining",
                "sec_num": "5.5"
            },
            {
                "text": "Second, fine-tuned models have a significant performance drop in the general language modeling tasks (Lambada and Wikitext). For example, the tuned model records 6x performance drop in the Wikitext compared to the best performing model. There are several factors in the continual pretraining that may contribute to this. As the tuned model uses the original GPT-2 vocabulary, it must use the fragmented general subwords to tokenize the chemistry terms available in our corpora. On the other hand, the tuned model starts with the suboptimal initialization from the general-domain language model (Gu et al., 2021) . This initialization may diverge the model in the optimization process that may not be recovered.",
                "cite_spans": [
                    {
                        "start": 594,
                        "end": 611,
                        "text": "(Gu et al., 2021)",
                        "ref_id": "BIBREF18"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Continual vs. From Scratch Pretraining",
                "sec_num": "5.5"
            },
            {
                "text": "We use several dimensions to describe the training efficiency, i.e., #FLOPs, throughput (speed), and memory. We compare these compute dimensions across the four model sizes described in the Table 2. The smallest (S) model has 59% FLOPs of the largest (XL) model, twice the speed (steps/s), 32% per device GPU memory savings, and 76% total parameter savings (see Figure 4) . With such compute budget, small (S) models only outperforms the XL model in 21% in-domain and 34% out-of-domain evaluation tasks. This suggests the importance of compute budget required in scaling foundation models.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 362,
                        "end": 371,
                        "text": "Figure 4)",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Training Efficiency",
                "sec_num": "5.6"
            },
            {
                "text": "In this paper, we collected and released 0.67TB of research publication data collected across 10+ sources for chemistry. We pretrained and released 25+ foundation models for chemistry. We rigorously analyzed model performance on 15+ indomain and out-of-domain tasks.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "6"
            },
            {
                "text": "AMiner ArnetMiner (AMiner) is a service that crawls research publications, performs profile extraction of scientists, models academic networks by integrating publication data from the existing libraries. For the experiments described in this work, we use a sub-sampled version of the data presented in the Open Academic Graph (OAG) version of the AMiner dataset, which originally consisted of more that 172M articles, with 18.5M chemistry-related abstracts.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Data Descriptions",
                "sec_num": null
            },
            {
                "text": "CORE COnnecting REpositories (CORE) (Pontika et al., 2016) is a large-scale aggregation system which provides an open access to the global network of scientific journals and publications. CORE currently contains more than 207M openaccess articles collected from over 10 thousand data providers, out of which more than 92M are open access full-text research papers. We sub-sampled the original collect into our chemistry-specific corpus consisting of more than 7M full-text articles.",
                "cite_spans": [
                    {
                        "start": 36,
                        "end": 58,
                        "text": "(Pontika et al., 2016)",
                        "ref_id": "BIBREF43"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Data Descriptions",
                "sec_num": null
            },
            {
                "text": "CORD-19 CORD-19 corpus contains COVID-19 (Cord19) and other coronavirus-related publications (e.g. SARS, MERS, etc.) from PubMed's PMC open access corpus, bioRxiv, and medRxiv pre-prints, in addition to COVID-19 articles maintained by the World Health Organization (WHO).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Data Descriptions",
                "sec_num": null
            },
            {
                "text": "MAG Microsoft Academic Graph (MAG) is a heterogeneous graph created by extracting knowledge from scholarly publications on the web (Wang et al., 2020a) . The data used in this work is a subsample from the OAG version of the MAG dataset, which originally consisted of > 208M articles, with 34M chemistry-related articles with abstracts.",
                "cite_spans": [
                    {
                        "start": 131,
                        "end": 151,
                        "text": "(Wang et al., 2020a)",
                        "ref_id": "BIBREF54"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Data Descriptions",
                "sec_num": null
            },
            {
                "text": "PubMed PubMed is a domain-specific data source that allows for search and retrieval of the biomedical and life sciences literature. It is maintained by the National Centre for Biotechnology Information (NCBI) at the U.S. National Library of Medicine (NLM). For this work we utilized the PubMed Central data provided in the Pile corpus . As presented in Table 3 the sub-sampled data consists of documents with more than 280K abstracts and 700K full text articles.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 353,
                        "end": 360,
                        "text": "Table 3",
                        "ref_id": "TABREF2"
                    }
                ],
                "eq_spans": [],
                "section": "A Data Descriptions",
                "sec_num": null
            },
            {
                "text": "The Semantic Scholar Open Research Corpus (S2ORC) (Lo et al., 2020 ) is a large academic corpus consisting of 81.1M documents. The data includes the metadata, abstracts, bibliographical references and full-text publications for over 8M open access research articles. In this work, we utilize the sub-sampled version of the original data specific to chemistry, which includes more than 10M abstracts.",
                "cite_spans": [
                    {
                        "start": 50,
                        "end": 66,
                        "text": "(Lo et al., 2020",
                        "ref_id": "BIBREF33"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "S2ORC",
                "sec_num": null
            },
            {
                "text": "WoS The Web of Science (WoS) is a multidiscipline citation database produced by the Institute of Scientific Information. The platform hosts over 171M records across various disciplines, which, when sub-sampled for our chemistry domain, rounded to more than 7M records with abstracts available.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "S2ORC",
                "sec_num": null
            },
            {
                "text": "HendrycksTest-Chemistry The Hendrycks Test (Hendrycks et al., 2020 ) is a large scale collection of multiple choice questions covering 57 subjects. In our experiments, we subsampled college chemistry (HT-CC) and high school chemistry (HT-HC). HT-CC contains 100 questions related to analytical, organic, inorganic, physical, etc. and HT-HC contains 203 questions related chemical reactions, ions, acids and bases, etc.",
                "cite_spans": [
                    {
                        "start": 43,
                        "end": 66,
                        "text": "(Hendrycks et al., 2020",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "ARC The ARC dataset contains 7,787 genuine grade-school level, science MCQs and is partitioned into a Challenge Set (ARC-C) and an Easy Set (ARC-E). Additionally, 14M science-related sentences are provided with relevant knowledge to answer the ARC questions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "SciQ The SciQ dataset (Welbl et al., 2017) contains 13,679 crowdsourced multiple-choice science exam questions about Physics, Chemistry and Biology, among others.",
                "cite_spans": [
                    {
                        "start": 22,
                        "end": 42,
                        "text": "(Welbl et al., 2017)",
                        "ref_id": "BIBREF56"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "OpenBookQA The OpenBookQA (Mihaylov et al., 2018) dataset consists of 5,957 multiple choice questions and 1,326 elementary-level science facts. The facts alone do not contain enough information to correctly answer the multiple choice questions, therefore the task is designed to evaluate systems beyond paraphrase matching.",
                "cite_spans": [
                    {
                        "start": 26,
                        "end": 49,
                        "text": "(Mihaylov et al., 2018)",
                        "ref_id": "BIBREF36"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "Pile PubMed Abstracts The Pile dataset contains 800GB of diverse text sources for benchmarking language models. We limit this task to only include abstracts from the Pile's PubMed collection. As this is framed as a language modeling task, we report word level perplexity.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "BoolQ BoolQ (Clark et al., 2019 ) is a reading comprehension dataset comprised of 16k real, naturally formed queries to the Google search engine with a yes or no answer. Each question-answer pair is accompanied by a Wikipedia article providing evidence to support the correct answer.",
                "cite_spans": [
                    {
                        "start": 12,
                        "end": 31,
                        "text": "(Clark et al., 2019",
                        "ref_id": "BIBREF10"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "CB Commitment Bank (CB) (De Marneffe et al., 2019 ) is a 3-way classification of textual entailment (true, false, unknown) from 1,200 short text segments where at least one sentence contains an embedded clause. The dataset contains passages from three sources: the Wall Street Journal, the British National Corpus, and Switchboard.",
                "cite_spans": [
                    {
                        "start": 28,
                        "end": 49,
                        "text": "Marneffe et al., 2019",
                        "ref_id": "BIBREF13"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "WIC The Word-in-Context dataset (WIC) (Pilehvar and Camacho-Collados, 2018) is a benchmark for evaluating context-sensitive word embeddings. The task is to classify if a target word has the same meaning in two context sentence.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "WSC The Winograd Schema Challenge (WSC) (Levesque et al., 2012) dataset is a collection of 804 sentences in which the task is to resolve coreferences.",
                "cite_spans": [
                    {
                        "start": 40,
                        "end": 63,
                        "text": "(Levesque et al., 2012)",
                        "ref_id": "BIBREF28"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "MathQA MathQA (Amini et al., 2019 ) is a dataset containing 37k multiple choice math word problems built from the existing dataset, AQuA (Ling et al., 2017) .",
                "cite_spans": [
                    {
                        "start": 14,
                        "end": 33,
                        "text": "(Amini et al., 2019",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 137,
                        "end": 156,
                        "text": "(Ling et al., 2017)",
                        "ref_id": "BIBREF30"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "PIQA The Physical Interactions: Question Answering (PIQA) (Bisk et al., 2020) benchmark dataset provides 21k questions about the physical world and plausible interactions encountered by humans. Annotators provided correct and incorrect answers to questions extracted from instructables.com, a website of instructions for completing many everyday tasks.",
                "cite_spans": [
                    {
                        "start": 58,
                        "end": 77,
                        "text": "(Bisk et al., 2020)",
                        "ref_id": "BIBREF5"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "PubMedQA The PubMedQA dataset ) is a collection of 273.5k biomedical research questions and related PubMed articles with yes/no/maybe answers.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "Lambada Lambada (Paperno et al., 2016) contains passages and target sentences from 5,325 novels collected from Book Corpus (Zhu et al., 2015) , and the goal is to predict the last word of the target sentence given the context passage. This task was designed to test genuine language understanding since accurate prediction of the final word would be improbable without the context passage.",
                "cite_spans": [
                    {
                        "start": 16,
                        "end": 38,
                        "text": "(Paperno et al., 2016)",
                        "ref_id": "BIBREF39"
                    },
                    {
                        "start": 123,
                        "end": 141,
                        "text": "(Zhu et al., 2015)",
                        "ref_id": "BIBREF58"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            },
            {
                "text": "WikiText The Wikitext benchmark (Merity et al., 2016 ) is a language modeling dataset of 29k articles from Wikipedia. Only articles classified as Good or Featured by Wikipedia editors are included since they are considered to be well written and neutral in language. All results are reported on Wikitext-2.",
                "cite_spans": [
                    {
                        "start": 32,
                        "end": 52,
                        "text": "(Merity et al., 2016",
                        "ref_id": "BIBREF35"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B Task Descriptions",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Publicly available clinical bert embeddings",
                "authors": [
                    {
                        "first": "Emily",
                        "middle": [],
                        "last": "Alsentzer",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "John",
                        "suffix": ""
                    },
                    {
                        "first": "Willie",
                        "middle": [],
                        "last": "Murphy",
                        "suffix": ""
                    },
                    {
                        "first": "Wei-Hung",
                        "middle": [],
                        "last": "Boag",
                        "suffix": ""
                    },
                    {
                        "first": "Di",
                        "middle": [],
                        "last": "Weng",
                        "suffix": ""
                    },
                    {
                        "first": "Tristan",
                        "middle": [],
                        "last": "Jin",
                        "suffix": ""
                    },
                    {
                        "first": "Matthew",
                        "middle": [],
                        "last": "Naumann",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Mcdermott",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1904.03323"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Emily Alsentzer, John R Murphy, Willie Boag, Wei- Hung Weng, Di Jin, Tristan Naumann, and Matthew McDermott. 2019. Publicly available clinical bert embeddings. arXiv preprint arXiv:1904.03323.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Mathqa: Towards interpretable math word problem solving with operation-based formalisms",
                "authors": [
                    {
                        "first": "Aida",
                        "middle": [],
                        "last": "Amini",
                        "suffix": ""
                    },
                    {
                        "first": "Saadia",
                        "middle": [],
                        "last": "Gabriel",
                        "suffix": ""
                    },
                    {
                        "first": "Peter",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "Rik",
                        "middle": [],
                        "last": "Koncel-Kedziorski",
                        "suffix": ""
                    },
                    {
                        "first": "Yejin",
                        "middle": [],
                        "last": "Choi",
                        "suffix": ""
                    },
                    {
                        "first": "Hannaneh",
                        "middle": [],
                        "last": "Hajishirzi",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1905.13319"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Aida Amini, Saadia Gabriel, Peter Lin, Rik Koncel- Kedziorski, Yejin Choi, and Hannaneh Hajishirzi. 2019. Mathqa: Towards interpretable math word problem solving with operation-based formalisms. arXiv preprint arXiv:1905.13319.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "GPT-NeoX: Large scale autoregressive language modeling in pytorch",
                "authors": [
                    {
                        "first": "Alex",
                        "middle": [],
                        "last": "Andonian",
                        "suffix": ""
                    },
                    {
                        "first": "Quentin",
                        "middle": [],
                        "last": "Anthony",
                        "suffix": ""
                    },
                    {
                        "first": "Stella",
                        "middle": [],
                        "last": "Biderman",
                        "suffix": ""
                    },
                    {
                        "first": "Sid",
                        "middle": [],
                        "last": "Black",
                        "suffix": ""
                    },
                    {
                        "first": "Preetham",
                        "middle": [],
                        "last": "Gali",
                        "suffix": ""
                    },
                    {
                        "first": "Leo",
                        "middle": [],
                        "last": "Gao",
                        "suffix": ""
                    },
                    {
                        "first": "Eric",
                        "middle": [],
                        "last": "Hallahan",
                        "suffix": ""
                    },
                    {
                        "first": "Josh",
                        "middle": [],
                        "last": "Levy-Kramer",
                        "suffix": ""
                    },
                    {
                        "first": "Connor",
                        "middle": [],
                        "last": "Leahy",
                        "suffix": ""
                    },
                    {
                        "first": "Lucas",
                        "middle": [],
                        "last": "Nestler",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Alex Andonian, Quentin Anthony, Stella Biderman, Sid Black, Preetham Gali, Leo Gao, Eric Hallahan, Josh Levy-Kramer, Connor Leahy, Lucas Nestler, Kip Parker, Michael Pieler, Shivanshu Purohit, Tri Songz, Phil Wang, and Samuel Weinbach. 2021. GPT-NeoX: Large scale autoregressive language modeling in py- torch.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Scibert: A pretrained language model for scientific text",
                "authors": [
                    {
                        "first": "Iz",
                        "middle": [],
                        "last": "Beltagy",
                        "suffix": ""
                    },
                    {
                        "first": "Kyle",
                        "middle": [],
                        "last": "Lo",
                        "suffix": ""
                    },
                    {
                        "first": "Arman",
                        "middle": [],
                        "last": "Cohan",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1903.10676"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Iz Beltagy, Kyle Lo, and Arman Cohan. 2019. Scibert: A pretrained language model for scientific text. arXiv preprint arXiv:1903.10676.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Piqa: Reasoning about physical commonsense in natural language",
                "authors": [
                    {
                        "first": "Yonatan",
                        "middle": [],
                        "last": "Bisk",
                        "suffix": ""
                    },
                    {
                        "first": "Rowan",
                        "middle": [],
                        "last": "Zellers",
                        "suffix": ""
                    },
                    {
                        "first": "Jianfeng",
                        "middle": [],
                        "last": "Gao",
                        "suffix": ""
                    },
                    {
                        "first": "Yejin",
                        "middle": [],
                        "last": "Choi",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the AAAI conference on artificial intelligence",
                "volume": "34",
                "issue": "",
                "pages": "7432--7439",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yonatan Bisk, Rowan Zellers, Jianfeng Gao, Yejin Choi, et al. 2020. Piqa: Reasoning about physical com- monsense in natural language. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 7432-7439.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Gpt-neox-20b: An open-source autoregressive language model",
                "authors": [
                    {
                        "first": "Sid",
                        "middle": [],
                        "last": "Black",
                        "suffix": ""
                    },
                    {
                        "first": "Stella",
                        "middle": [],
                        "last": "Biderman",
                        "suffix": ""
                    },
                    {
                        "first": "Eric",
                        "middle": [],
                        "last": "Hallahan",
                        "suffix": ""
                    },
                    {
                        "first": "Quentin",
                        "middle": [],
                        "last": "Anthony",
                        "suffix": ""
                    },
                    {
                        "first": "Leo",
                        "middle": [],
                        "last": "Gao",
                        "suffix": ""
                    },
                    {
                        "first": "Laurence",
                        "middle": [],
                        "last": "Golding",
                        "suffix": ""
                    },
                    {
                        "first": "Horace",
                        "middle": [],
                        "last": "He",
                        "suffix": ""
                    },
                    {
                        "first": "Connor",
                        "middle": [],
                        "last": "Leahy",
                        "suffix": ""
                    },
                    {
                        "first": "Kyle",
                        "middle": [],
                        "last": "Mcdonell",
                        "suffix": ""
                    },
                    {
                        "first": "Jason",
                        "middle": [],
                        "last": "Phang",
                        "suffix": ""
                    }
                ],
                "year": 2022,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, et al. 2022. Gpt-neox-20b: An open-source autoregressive language model.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "On the opportunities and risks of foundation models",
                "authors": [
                    {
                        "first": "Rishi",
                        "middle": [],
                        "last": "Bommasani",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Drew",
                        "suffix": ""
                    },
                    {
                        "first": "Ehsan",
                        "middle": [],
                        "last": "Hudson",
                        "suffix": ""
                    },
                    {
                        "first": "Russ",
                        "middle": [],
                        "last": "Adeli",
                        "suffix": ""
                    },
                    {
                        "first": "Simran",
                        "middle": [],
                        "last": "Altman",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Arora",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Sydney Von Arx",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Michael",
                        "suffix": ""
                    },
                    {
                        "first": "Jeannette",
                        "middle": [],
                        "last": "Bernstein",
                        "suffix": ""
                    },
                    {
                        "first": "Antoine",
                        "middle": [],
                        "last": "Bohg",
                        "suffix": ""
                    },
                    {
                        "first": "Emma",
                        "middle": [],
                        "last": "Bosselut",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Brunskill",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2108.07258"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Rishi Bommasani, Drew A Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S Bernstein, Jeannette Bohg, Antoine Bosse- lut, Emma Brunskill, et al. 2021. On the opportuni- ties and risks of foundation models. arXiv preprint arXiv:2108.07258.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Language models are few-shot learners",
                "authors": [
                    {
                        "first": "Tom",
                        "middle": [],
                        "last": "Brown",
                        "suffix": ""
                    },
                    {
                        "first": "Benjamin",
                        "middle": [],
                        "last": "Mann",
                        "suffix": ""
                    },
                    {
                        "first": "Nick",
                        "middle": [],
                        "last": "Ryder",
                        "suffix": ""
                    },
                    {
                        "first": "Melanie",
                        "middle": [],
                        "last": "Subbiah",
                        "suffix": ""
                    },
                    {
                        "first": "Jared",
                        "middle": [
                            "D"
                        ],
                        "last": "Kaplan",
                        "suffix": ""
                    },
                    {
                        "first": "Prafulla",
                        "middle": [],
                        "last": "Dhariwal",
                        "suffix": ""
                    },
                    {
                        "first": "Arvind",
                        "middle": [],
                        "last": "Neelakantan",
                        "suffix": ""
                    },
                    {
                        "first": "Pranav",
                        "middle": [],
                        "last": "Shyam",
                        "suffix": ""
                    },
                    {
                        "first": "Girish",
                        "middle": [],
                        "last": "Sastry",
                        "suffix": ""
                    },
                    {
                        "first": "Amanda",
                        "middle": [],
                        "last": "Askell",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Advances in neural information processing systems",
                "volume": "33",
                "issue": "",
                "pages": "1877--1901",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877-1901.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Quantifying memorization across neural language models",
                "authors": [
                    {
                        "first": "Nicholas",
                        "middle": [],
                        "last": "Carlini",
                        "suffix": ""
                    },
                    {
                        "first": "Daphne",
                        "middle": [],
                        "last": "Ippolito",
                        "suffix": ""
                    },
                    {
                        "first": "Matthew",
                        "middle": [],
                        "last": "Jagielski",
                        "suffix": ""
                    },
                    {
                        "first": "Katherine",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "Florian",
                        "middle": [],
                        "last": "Tramer",
                        "suffix": ""
                    },
                    {
                        "first": "Chiyuan",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    }
                ],
                "year": 2022,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2202.07646"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Nicholas Carlini, Daphne Ippolito, Matthew Jagielski, Katherine Lee, Florian Tramer, and Chiyuan Zhang. 2022. Quantifying memorization across neural lan- guage models. arXiv preprint arXiv:2202.07646.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Boolq: Exploring the surprising difficulty of natural yes/no questions",
                "authors": [
                    {
                        "first": "Christopher",
                        "middle": [],
                        "last": "Clark",
                        "suffix": ""
                    },
                    {
                        "first": "Kenton",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "Ming-Wei",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    },
                    {
                        "first": "Tom",
                        "middle": [],
                        "last": "Kwiatkowski",
                        "suffix": ""
                    },
                    {
                        "first": "Michael",
                        "middle": [],
                        "last": "Collins",
                        "suffix": ""
                    },
                    {
                        "first": "Kristina",
                        "middle": [],
                        "last": "Toutanova",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "NAACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina Toutanova. 2019. Boolq: Exploring the surprising difficulty of natural yes/no questions. In NAACL.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Think you have solved question answering? try arc, the ai2 reasoning challenge",
                "authors": [
                    {
                        "first": "Peter",
                        "middle": [],
                        "last": "Clark",
                        "suffix": ""
                    },
                    {
                        "first": "Isaac",
                        "middle": [],
                        "last": "Cowhey",
                        "suffix": ""
                    },
                    {
                        "first": "Oren",
                        "middle": [],
                        "last": "Etzioni",
                        "suffix": ""
                    },
                    {
                        "first": "Tushar",
                        "middle": [],
                        "last": "Khot",
                        "suffix": ""
                    },
                    {
                        "first": "Ashish",
                        "middle": [],
                        "last": "Sabharwal",
                        "suffix": ""
                    },
                    {
                        "first": "Carissa",
                        "middle": [],
                        "last": "Schoenick",
                        "suffix": ""
                    },
                    {
                        "first": "Oyvind",
                        "middle": [],
                        "last": "Tafjord",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1803.05457"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. 2018. Think you have solved question an- swering? try arc, the ai2 reasoning challenge. arXiv preprint arXiv:1803.05457.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "The commitmentbank: Investigating projection in naturally occurring discourse",
                "authors": [
                    {
                        "first": "Marie-Catherine De",
                        "middle": [],
                        "last": "Marneffe",
                        "suffix": ""
                    },
                    {
                        "first": "Mandy",
                        "middle": [],
                        "last": "Simons",
                        "suffix": ""
                    },
                    {
                        "first": "Judith",
                        "middle": [],
                        "last": "Tonhauser",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "proceedings of Sinn und Bedeutung",
                "volume": "23",
                "issue": "",
                "pages": "107--124",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Marie-Catherine De Marneffe, Mandy Simons, and Ju- dith Tonhauser. 2019. The commitmentbank: Inves- tigating projection in naturally occurring discourse. In proceedings of Sinn und Bedeutung, volume 23, pages 107-124.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Bert: Pre-training of deep bidirectional transformers for language understanding",
                "authors": [
                    {
                        "first": "Jacob",
                        "middle": [],
                        "last": "Devlin",
                        "suffix": ""
                    },
                    {
                        "first": "Ming-Wei",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    },
                    {
                        "first": "Kenton",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "Kristina",
                        "middle": [],
                        "last": "Toutanova",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1810.04805"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understand- ing. arXiv preprint arXiv:1810.04805.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "Anchored correlation explanation: Topic modeling with minimal domain knowledge",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Ryan",
                        "suffix": ""
                    },
                    {
                        "first": "Kyle",
                        "middle": [],
                        "last": "Gallagher",
                        "suffix": ""
                    },
                    {
                        "first": "David",
                        "middle": [],
                        "last": "Reing",
                        "suffix": ""
                    },
                    {
                        "first": "Greg",
                        "middle": [
                            "Ver"
                        ],
                        "last": "Kale",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Steeg",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Transactions of the Association for Computational Linguistics",
                "volume": "5",
                "issue": "",
                "pages": "529--542",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ryan J Gallagher, Kyle Reing, David Kale, and Greg Ver Steeg. 2017. Anchored correlation explanation: Topic modeling with minimal domain knowledge. Transactions of the Association for Computational Linguistics, 5:529-542.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "The pile: An 800gb dataset of diverse text for language modeling",
                "authors": [
                    {
                        "first": "Leo",
                        "middle": [],
                        "last": "Gao",
                        "suffix": ""
                    },
                    {
                        "first": "Stella",
                        "middle": [],
                        "last": "Biderman",
                        "suffix": ""
                    },
                    {
                        "first": "Sid",
                        "middle": [],
                        "last": "Black",
                        "suffix": ""
                    },
                    {
                        "first": "Laurence",
                        "middle": [],
                        "last": "Golding",
                        "suffix": ""
                    },
                    {
                        "first": "Travis",
                        "middle": [],
                        "last": "Hoppe",
                        "suffix": ""
                    },
                    {
                        "first": "Charles",
                        "middle": [],
                        "last": "Foster",
                        "suffix": ""
                    },
                    {
                        "first": "Jason",
                        "middle": [],
                        "last": "Phang",
                        "suffix": ""
                    },
                    {
                        "first": "Horace",
                        "middle": [],
                        "last": "He",
                        "suffix": ""
                    },
                    {
                        "first": "Anish",
                        "middle": [],
                        "last": "Thite",
                        "suffix": ""
                    },
                    {
                        "first": "Noa",
                        "middle": [],
                        "last": "Nabeshima",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2101.00027"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Leo Gao, Stella Biderman, Sid Black, Laurence Gold- ing, Travis Hoppe, Charles Foster, Jason Phang, Ho- race He, Anish Thite, Noa Nabeshima, et al. 2020. The pile: An 800gb dataset of diverse text for lan- guage modeling. arXiv preprint arXiv:2101.00027.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Domain-specific language model pretraining for biomedical natural language processing",
                "authors": [
                    {
                        "first": "Yu",
                        "middle": [],
                        "last": "Gu",
                        "suffix": ""
                    },
                    {
                        "first": "Robert",
                        "middle": [],
                        "last": "Tinn",
                        "suffix": ""
                    },
                    {
                        "first": "Hao",
                        "middle": [],
                        "last": "Cheng",
                        "suffix": ""
                    },
                    {
                        "first": "Michael",
                        "middle": [],
                        "last": "Lucas",
                        "suffix": ""
                    },
                    {
                        "first": "Naoto",
                        "middle": [],
                        "last": "Usuyama",
                        "suffix": ""
                    },
                    {
                        "first": "Xiaodong",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "Tristan",
                        "middle": [],
                        "last": "Naumann",
                        "suffix": ""
                    },
                    {
                        "first": "Jianfeng",
                        "middle": [],
                        "last": "Gao",
                        "suffix": ""
                    },
                    {
                        "first": "Hoifung",
                        "middle": [],
                        "last": "Poon",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "ACM Transactions on Computing for Healthcare",
                "volume": "3",
                "issue": "1",
                "pages": "1--23",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. 2021. Domain-specific lan- guage model pretraining for biomedical natural lan- guage processing. ACM Transactions on Computing for Healthcare (HEALTH), 3(1):1-23.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Automated chemical reaction extraction from scientific literature",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Jiang Guo",
                        "suffix": ""
                    },
                    {
                        "first": "Hanyu",
                        "middle": [],
                        "last": "Santiago Ibanez-Lopez",
                        "suffix": ""
                    },
                    {
                        "first": "Victor",
                        "middle": [],
                        "last": "Gao",
                        "suffix": ""
                    },
                    {
                        "first": "Connor",
                        "middle": [
                            "W"
                        ],
                        "last": "Quach",
                        "suffix": ""
                    },
                    {
                        "first": "Klavs",
                        "middle": [
                            "F"
                        ],
                        "last": "Coley",
                        "suffix": ""
                    },
                    {
                        "first": "Regina",
                        "middle": [],
                        "last": "Jensen",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Barzilay",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "Journal of Chemical Information and Modeling",
                "volume": "0",
                "issue": "0",
                "pages": "",
                "other_ids": {
                    "DOI": [
                        "10.1021/acs.jcim.1c00284"
                    ],
                    "PMID": [
                        "34115937"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Jiang Guo, A. Santiago Ibanez-Lopez, Hanyu Gao, Vic- tor Quach, Connor W. Coley, Klavs F. Jensen, and Regina Barzilay. 2021. Automated chemical reaction extraction from scientific literature. Journal of Chem- ical Information and Modeling, 0(0):null. PMID: 34115937.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2020. Measuring massive multitask language understanding",
                "authors": [
                    {
                        "first": "Dan",
                        "middle": [],
                        "last": "Hendrycks",
                        "suffix": ""
                    },
                    {
                        "first": "Collin",
                        "middle": [],
                        "last": "Burns",
                        "suffix": ""
                    },
                    {
                        "first": "Steven",
                        "middle": [],
                        "last": "Basart",
                        "suffix": ""
                    },
                    {
                        "first": "Andy",
                        "middle": [],
                        "last": "Zou",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2009.03300"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2020. Measuring massive multitask language under- standing. arXiv preprint arXiv:2009.03300.",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "Perceiver: General perception with iterative attention",
                "authors": [
                    {
                        "first": "Andrew",
                        "middle": [],
                        "last": "Jaegle",
                        "suffix": ""
                    },
                    {
                        "first": "Felix",
                        "middle": [],
                        "last": "Gimeno",
                        "suffix": ""
                    },
                    {
                        "first": "Andy",
                        "middle": [],
                        "last": "Brock",
                        "suffix": ""
                    },
                    {
                        "first": "Oriol",
                        "middle": [],
                        "last": "Vinyals",
                        "suffix": ""
                    },
                    {
                        "first": "Andrew",
                        "middle": [],
                        "last": "Zisserman",
                        "suffix": ""
                    },
                    {
                        "first": "Joao",
                        "middle": [],
                        "last": "Carreira",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "International Conference on Machine Learning",
                "volume": "",
                "issue": "",
                "pages": "4651--4664",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Andrew Jaegle, Felix Gimeno, Andy Brock, Oriol Vinyals, Andrew Zisserman, and Joao Carreira. 2021. Perceiver: General perception with iterative attention. In International Conference on Machine Learning, pages 4651-4664. PMLR.",
                "links": null
            },
            "BIBREF22": {
                "ref_id": "b22",
                "title": "Pubmedqa: A dataset for biomedical research question answering",
                "authors": [
                    {
                        "first": "Qiao",
                        "middle": [],
                        "last": "Jin",
                        "suffix": ""
                    },
                    {
                        "first": "Bhuwan",
                        "middle": [],
                        "last": "Dhingra",
                        "suffix": ""
                    },
                    {
                        "first": "Zhengping",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [],
                        "last": "William",
                        "suffix": ""
                    },
                    {
                        "first": "Xinghua",
                        "middle": [],
                        "last": "Cohen",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Lu",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1909.06146"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Qiao Jin, Bhuwan Dhingra, Zhengping Liu, William W Cohen, and Xinghua Lu. 2019. Pubmedqa: A dataset for biomedical research question answering. arXiv preprint arXiv:1909.06146.",
                "links": null
            },
            "BIBREF23": {
                "ref_id": "b23",
                "title": "Bioelectra: pretrained biomedical text encoder using discriminators",
                "authors": [
                    {
                        "first": "Bhuvana",
                        "middle": [],
                        "last": "Kamal Kanakarajan",
                        "suffix": ""
                    },
                    {
                        "first": "Malaikannan",
                        "middle": [],
                        "last": "Kundumani",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Sankarasubbu",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "Proceedings of the 20th Workshop on Biomedical Language Processing",
                "volume": "",
                "issue": "",
                "pages": "143--154",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kamal Kanakarajan, Bhuvana Kundumani, and Malaikannan Sankarasubbu. 2021. Bioelectra: pre- trained biomedical text encoder using discriminators. In Proceedings of the 20th Workshop on Biomedical Language Processing, pages 143-154.",
                "links": null
            },
            "BIBREF24": {
                "ref_id": "b24",
                "title": "and Dario Amodei. 2020. Scaling laws for neural language models",
                "authors": [
                    {
                        "first": "Jared",
                        "middle": [],
                        "last": "Kaplan",
                        "suffix": ""
                    },
                    {
                        "first": "Sam",
                        "middle": [],
                        "last": "Mccandlish",
                        "suffix": ""
                    },
                    {
                        "first": "Tom",
                        "middle": [],
                        "last": "Henighan",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Tom",
                        "suffix": ""
                    },
                    {
                        "first": "Benjamin",
                        "middle": [],
                        "last": "Brown",
                        "suffix": ""
                    },
                    {
                        "first": "Rewon",
                        "middle": [],
                        "last": "Chess",
                        "suffix": ""
                    },
                    {
                        "first": "Scott",
                        "middle": [],
                        "last": "Child",
                        "suffix": ""
                    },
                    {
                        "first": "Alec",
                        "middle": [],
                        "last": "Gray",
                        "suffix": ""
                    },
                    {
                        "first": "Jeffrey",
                        "middle": [],
                        "last": "Radford",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2001.08361"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. 2020. Scaling laws for neural language models. arXiv preprint arXiv:2001.08361.",
                "links": null
            },
            "BIBREF25": {
                "ref_id": "b25",
                "title": "Mixout: Effective regularization to finetune large-scale pretrained language models",
                "authors": [
                    {
                        "first": "Cheolhyoung",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "Kyunghyun",
                        "middle": [],
                        "last": "Cho",
                        "suffix": ""
                    },
                    {
                        "first": "Wanmo",
                        "middle": [],
                        "last": "Kang",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1909.11299"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Cheolhyoung Lee, Kyunghyun Cho, and Wanmo Kang. 2019. Mixout: Effective regularization to fine- tune large-scale pretrained language models. arXiv preprint arXiv:1909.11299.",
                "links": null
            },
            "BIBREF26": {
                "ref_id": "b26",
                "title": "Biobert: a pre-trained biomedical language representation model for biomedical text mining",
                "authors": [
                    {
                        "first": "Jinhyuk",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "Wonjin",
                        "middle": [],
                        "last": "Yoon",
                        "suffix": ""
                    },
                    {
                        "first": "Sungdong",
                        "middle": [],
                        "last": "Kim",
                        "suffix": ""
                    },
                    {
                        "first": "Donghyeon",
                        "middle": [],
                        "last": "Kim",
                        "suffix": ""
                    },
                    {
                        "first": "Sunkyu",
                        "middle": [],
                        "last": "Kim",
                        "suffix": ""
                    },
                    {
                        "first": "Chan",
                        "middle": [],
                        "last": "Ho So",
                        "suffix": ""
                    },
                    {
                        "first": "Jaewoo",
                        "middle": [],
                        "last": "Kang",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Bioinformatics",
                "volume": "36",
                "issue": "4",
                "pages": "1234--1240",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jaewoo Kang. 2020. Biobert: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics, 36(4):1234-1240.",
                "links": null
            },
            "BIBREF27": {
                "ref_id": "b27",
                "title": "Chris Callison-Burch, and Nicholas Carlini. 2021. Deduplicating training data makes language models better",
                "authors": [
                    {
                        "first": "Katherine",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "Daphne",
                        "middle": [],
                        "last": "Ippolito",
                        "suffix": ""
                    },
                    {
                        "first": "Andrew",
                        "middle": [],
                        "last": "Nystrom",
                        "suffix": ""
                    },
                    {
                        "first": "Chiyuan",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "Douglas",
                        "middle": [],
                        "last": "Eck",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2107.06499"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Katherine Lee, Daphne Ippolito, Andrew Nystrom, Chiyuan Zhang, Douglas Eck, Chris Callison-Burch, and Nicholas Carlini. 2021. Deduplicating training data makes language models better. arXiv preprint arXiv:2107.06499.",
                "links": null
            },
            "BIBREF28": {
                "ref_id": "b28",
                "title": "The winograd schema challenge",
                "authors": [
                    {
                        "first": "Hector",
                        "middle": [
                            "J"
                        ],
                        "last": "Levesque",
                        "suffix": ""
                    },
                    {
                        "first": "Ernest",
                        "middle": [],
                        "last": "Davis",
                        "suffix": ""
                    },
                    {
                        "first": "Leora",
                        "middle": [],
                        "last": "Morgenstern",
                        "suffix": ""
                    }
                ],
                "year": 2012,
                "venue": "",
                "volume": "12",
                "issue": "",
                "pages": "552--561",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hector J. Levesque, Ernest Davis, and Leora Morgen- stern. 2012. The winograd schema challenge. KR'12, page 552-561. AAAI Press.",
                "links": null
            },
            "BIBREF29": {
                "ref_id": "b29",
                "title": "Pretrained language models for biomedical and clinical tasks: Understanding and extending the state-of-the-art",
                "authors": [
                    {
                        "first": "Patrick",
                        "middle": [],
                        "last": "Lewis",
                        "suffix": ""
                    },
                    {
                        "first": "Myle",
                        "middle": [],
                        "last": "Ott",
                        "suffix": ""
                    },
                    {
                        "first": "Jingfei",
                        "middle": [],
                        "last": "Du",
                        "suffix": ""
                    },
                    {
                        "first": "Veselin",
                        "middle": [],
                        "last": "Stoyanov",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 3rd Clinical Natural Language Processing Workshop",
                "volume": "",
                "issue": "",
                "pages": "146--157",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Patrick Lewis, Myle Ott, Jingfei Du, and Veselin Stoy- anov. 2020. Pretrained language models for biomedi- cal and clinical tasks: Understanding and extending the state-of-the-art. In Proceedings of the 3rd Clini- cal Natural Language Processing Workshop, pages 146-157.",
                "links": null
            },
            "BIBREF30": {
                "ref_id": "b30",
                "title": "Program induction by rationale generation: Learning to solve and explain algebraic word problems",
                "authors": [
                    {
                        "first": "Wang",
                        "middle": [],
                        "last": "Ling",
                        "suffix": ""
                    },
                    {
                        "first": "Dani",
                        "middle": [],
                        "last": "Yogatama",
                        "suffix": ""
                    },
                    {
                        "first": "Chris",
                        "middle": [],
                        "last": "Dyer",
                        "suffix": ""
                    },
                    {
                        "first": "Phil",
                        "middle": [],
                        "last": "Blunsom",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1705.04146"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blun- som. 2017. Program induction by rationale genera- tion: Learning to solve and explain algebraic word problems. arXiv preprint arXiv:1705.04146.",
                "links": null
            },
            "BIBREF31": {
                "ref_id": "b31",
                "title": "Oag-bert: Pretrain heterogeneous entity-augmented academic language models",
                "authors": [
                    {
                        "first": "Xiao",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "Xingjian",
                        "middle": [],
                        "last": "Da Yin",
                        "suffix": ""
                    },
                    {
                        "first": "Kai",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "Kan",
                        "middle": [],
                        "last": "Su",
                        "suffix": ""
                    },
                    {
                        "first": "Hongxia",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "Jie",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Tang",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2103.02410"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Xiao Liu, Da Yin, Xingjian Zhang, Kai Su, Kan Wu, Hongxia Yang, and Jie Tang. 2021. Oag-bert: Pre- train heterogeneous entity-augmented academic lan- guage models. arXiv preprint arXiv:2103.02410.",
                "links": null
            },
            "BIBREF32": {
                "ref_id": "b32",
                "title": "Roberta: A robustly optimized bert pretraining approach",
                "authors": [
                    {
                        "first": "Yinhan",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "Myle",
                        "middle": [],
                        "last": "Ott",
                        "suffix": ""
                    },
                    {
                        "first": "Naman",
                        "middle": [],
                        "last": "Goyal",
                        "suffix": ""
                    },
                    {
                        "first": "Jingfei",
                        "middle": [],
                        "last": "Du",
                        "suffix": ""
                    },
                    {
                        "first": "Mandar",
                        "middle": [],
                        "last": "Joshi",
                        "suffix": ""
                    },
                    {
                        "first": "Danqi",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Omer",
                        "middle": [],
                        "last": "Levy",
                        "suffix": ""
                    },
                    {
                        "first": "Mike",
                        "middle": [],
                        "last": "Lewis",
                        "suffix": ""
                    },
                    {
                        "first": "Luke",
                        "middle": [],
                        "last": "Zettlemoyer",
                        "suffix": ""
                    },
                    {
                        "first": "Veselin",
                        "middle": [],
                        "last": "Stoyanov",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1907.11692"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Man- dar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining ap- proach. arXiv preprint arXiv:1907.11692.",
                "links": null
            },
            "BIBREF33": {
                "ref_id": "b33",
                "title": "S2orc: The semantic scholar open research corpus",
                "authors": [
                    {
                        "first": "Kyle",
                        "middle": [],
                        "last": "Lo",
                        "suffix": ""
                    },
                    {
                        "first": "Lucy",
                        "middle": [
                            "Lu"
                        ],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Neumann",
                        "suffix": ""
                    },
                    {
                        "first": "Rodney",
                        "middle": [
                            "Michael"
                        ],
                        "last": "Kinney",
                        "suffix": ""
                    },
                    {
                        "first": "Daniel",
                        "middle": [
                            "S"
                        ],
                        "last": "Weld",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "ACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kyle Lo, Lucy Lu Wang, Mark Neumann, Rod- ney Michael Kinney, and Daniel S. Weld. 2020. S2orc: The semantic scholar open research corpus. In ACL.",
                "links": null
            },
            "BIBREF34": {
                "ref_id": "b34",
                "title": "Time waits for no one! analysis and challenges of temporal misalignment",
                "authors": [
                    {
                        "first": "Kelvin",
                        "middle": [],
                        "last": "Luu",
                        "suffix": ""
                    },
                    {
                        "first": "Daniel",
                        "middle": [],
                        "last": "Khashabi",
                        "suffix": ""
                    },
                    {
                        "first": "Suchin",
                        "middle": [],
                        "last": "Gururangan",
                        "suffix": ""
                    },
                    {
                        "first": "Karishma",
                        "middle": [],
                        "last": "Mandyam",
                        "suffix": ""
                    },
                    {
                        "first": "Noah A",
                        "middle": [],
                        "last": "Smith",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2111.07408"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Kelvin Luu, Daniel Khashabi, Suchin Gururangan, Kar- ishma Mandyam, and Noah A Smith. 2021. Time waits for no one! analysis and challenges of temporal misalignment. arXiv preprint arXiv:2111.07408.",
                "links": null
            },
            "BIBREF35": {
                "ref_id": "b35",
                "title": "Pointer sentinel mixture models",
                "authors": [
                    {
                        "first": "Stephen",
                        "middle": [],
                        "last": "Merity",
                        "suffix": ""
                    },
                    {
                        "first": "Caiming",
                        "middle": [],
                        "last": "Xiong",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [],
                        "last": "Bradbury",
                        "suffix": ""
                    },
                    {
                        "first": "Richard",
                        "middle": [],
                        "last": "Socher",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1609.07843"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. 2016. Pointer sentinel mixture mod- els. arXiv preprint arXiv:1609.07843.",
                "links": null
            },
            "BIBREF36": {
                "ref_id": "b36",
                "title": "Can a suit of armor conduct electricity? a new dataset for open book question answering",
                "authors": [
                    {
                        "first": "Todor",
                        "middle": [],
                        "last": "Mihaylov",
                        "suffix": ""
                    },
                    {
                        "first": "Peter",
                        "middle": [],
                        "last": "Clark",
                        "suffix": ""
                    },
                    {
                        "first": "Tushar",
                        "middle": [],
                        "last": "Khot",
                        "suffix": ""
                    },
                    {
                        "first": "Ashish",
                        "middle": [],
                        "last": "Sabharwal",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1809.02789"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Todor Mihaylov, Peter Clark, Tushar Khot, and Ashish Sabharwal. 2018. Can a suit of armor conduct elec- tricity? a new dataset for open book question answer- ing. arXiv preprint arXiv:1809.02789.",
                "links": null
            },
            "BIBREF37": {
                "ref_id": "b37",
                "title": "Electramed: a new pre-trained language representation model for biomedical nlp",
                "authors": [
                    {
                        "first": "Giacomo",
                        "middle": [],
                        "last": "Miolo",
                        "suffix": ""
                    },
                    {
                        "first": "Giulio",
                        "middle": [],
                        "last": "Mantoan",
                        "suffix": ""
                    },
                    {
                        "first": "Carlotta",
                        "middle": [],
                        "last": "Orsenigo",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2104.09585"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Giacomo Miolo, Giulio Mantoan, and Carlotta Orsenigo. 2021. Electramed: a new pre-trained language repre- sentation model for biomedical nlp. arXiv preprint arXiv:2104.09585.",
                "links": null
            },
            "BIBREF38": {
                "ref_id": "b38",
                "title": "Benchmarking for biomedical natural language processing tasks with a domain specific albert",
                "authors": [
                    {
                        "first": "Usman",
                        "middle": [],
                        "last": "Naseem",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Adam",
                        "suffix": ""
                    },
                    {
                        "first": "Matloob",
                        "middle": [],
                        "last": "Dunn",
                        "suffix": ""
                    },
                    {
                        "first": "Jinman",
                        "middle": [],
                        "last": "Khushi",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Kim",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2107.04374"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Usman Naseem, Adam G Dunn, Matloob Khushi, and Jinman Kim. 2021. Benchmarking for biomedical natural language processing tasks with a domain spe- cific albert. arXiv preprint arXiv:2107.04374. OAG. https://www.microsoft.com/en- us/research/project/open-academic-graph/.",
                "links": null
            },
            "BIBREF39": {
                "ref_id": "b39",
                "title": "The lambada dataset: Word prediction requiring a broad discourse context",
                "authors": [
                    {
                        "first": "Denis",
                        "middle": [],
                        "last": "Paperno",
                        "suffix": ""
                    },
                    {
                        "first": "Germ\u00e1n",
                        "middle": [],
                        "last": "Kruszewski",
                        "suffix": ""
                    },
                    {
                        "first": "Angeliki",
                        "middle": [],
                        "last": "Lazaridou",
                        "suffix": ""
                    },
                    {
                        "first": "Ngoc",
                        "middle": [],
                        "last": "Quan",
                        "suffix": ""
                    },
                    {
                        "first": "Raffaella",
                        "middle": [],
                        "last": "Pham",
                        "suffix": ""
                    },
                    {
                        "first": "Sandro",
                        "middle": [],
                        "last": "Bernardi",
                        "suffix": ""
                    },
                    {
                        "first": "Marco",
                        "middle": [],
                        "last": "Pezzelle",
                        "suffix": ""
                    },
                    {
                        "first": "Gemma",
                        "middle": [],
                        "last": "Baroni",
                        "suffix": ""
                    },
                    {
                        "first": "Raquel",
                        "middle": [],
                        "last": "Boleda",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Fern\u00e1ndez",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1606.06031"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Denis Paperno, Germ\u00e1n Kruszewski, Angeliki Lazari- dou, Quan Ngoc Pham, Raffaella Bernardi, Sandro Pezzelle, Marco Baroni, Gemma Boleda, and Raquel Fern\u00e1ndez. 2016. The lambada dataset: Word pre- diction requiring a broad discourse context. arXiv preprint arXiv:1606.06031.",
                "links": null
            },
            "BIBREF40": {
                "ref_id": "b40",
                "title": "Transfer learning in biomedical natural language processing: an evaluation of bert and elmo on ten benchmarking datasets",
                "authors": [
                    {
                        "first": "Yifan",
                        "middle": [],
                        "last": "Peng",
                        "suffix": ""
                    },
                    {
                        "first": "Shankai",
                        "middle": [],
                        "last": "Yan",
                        "suffix": ""
                    },
                    {
                        "first": "Zhiyong",
                        "middle": [],
                        "last": "Lu",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1906.05474"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Yifan Peng, Shankai Yan, and Zhiyong Lu. 2019. Trans- fer learning in biomedical natural language process- ing: an evaluation of bert and elmo on ten bench- marking datasets. arXiv preprint arXiv:1906.05474.",
                "links": null
            },
            "BIBREF41": {
                "ref_id": "b41",
                "title": "Erol Bahadroglu, Alec Peltekian, and Gr\u00e9goire Altan-Bonnet. 2021. Scifive: a text-to-text transformer model for biomedical literature",
                "authors": [
                    {
                        "first": "James",
                        "middle": [
                            "T"
                        ],
                        "last": "Long N Phan",
                        "suffix": ""
                    },
                    {
                        "first": "Hieu",
                        "middle": [],
                        "last": "Anibal",
                        "suffix": ""
                    },
                    {
                        "first": "Shaurya",
                        "middle": [],
                        "last": "Tran",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Chanana",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2106.03598"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Long N Phan, James T Anibal, Hieu Tran, Shaurya Chanana, Erol Bahadroglu, Alec Peltekian, and Gr\u00e9- goire Altan-Bonnet. 2021. Scifive: a text-to-text transformer model for biomedical literature. arXiv preprint arXiv:2106.03598.",
                "links": null
            },
            "BIBREF42": {
                "ref_id": "b42",
                "title": "Wic: the word-in-context dataset for evaluating context-sensitive meaning representations",
                "authors": [
                    {
                        "first": "Mohammad",
                        "middle": [],
                        "last": "Taher Pilehvar",
                        "suffix": ""
                    },
                    {
                        "first": "Jose",
                        "middle": [],
                        "last": "Camacho-Collados",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1808.09121"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Mohammad Taher Pilehvar and Jose Camacho-Collados. 2018. Wic: the word-in-context dataset for evaluat- ing context-sensitive meaning representations. arXiv preprint arXiv:1808.09121.",
                "links": null
            },
            "BIBREF43": {
                "ref_id": "b43",
                "title": "Developing infrastructure to support closer collaboration of aggregators with open repositories",
                "authors": [
                    {
                        "first": "Nancy",
                        "middle": [],
                        "last": "Pontika",
                        "suffix": ""
                    },
                    {
                        "first": "Petr",
                        "middle": [],
                        "last": "Knoth",
                        "suffix": ""
                    },
                    {
                        "first": "Matteo",
                        "middle": [],
                        "last": "Cancellieri",
                        "suffix": ""
                    },
                    {
                        "first": "Samuel",
                        "middle": [],
                        "last": "Pearce",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "LIBER Quarterly",
                "volume": "25",
                "issue": "4",
                "pages": "172--188",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Nancy Pontika, Petr Knoth, Matteo Cancellieri, and Samuel Pearce. 2016. Developing infrastructure to support closer collaboration of aggregators with open repositories. LIBER Quarterly, 25(4):172-188.",
                "links": null
            },
            "BIBREF44": {
                "ref_id": "b44",
                "title": "Language models are unsupervised multitask learners",
                "authors": [
                    {
                        "first": "Alec",
                        "middle": [],
                        "last": "Radford",
                        "suffix": ""
                    },
                    {
                        "first": "Jeffrey",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "Rewon",
                        "middle": [],
                        "last": "Child",
                        "suffix": ""
                    },
                    {
                        "first": "David",
                        "middle": [],
                        "last": "Luan",
                        "suffix": ""
                    },
                    {
                        "first": "Dario",
                        "middle": [],
                        "last": "Amodei",
                        "suffix": ""
                    },
                    {
                        "first": "Ilya",
                        "middle": [],
                        "last": "Sutskever",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "OpenAI blog",
                "volume": "1",
                "issue": "8",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9.",
                "links": null
            },
            "BIBREF45": {
                "ref_id": "b45",
                "title": "Scaling language models: Methods, analysis & insights from training gopher",
                "authors": [
                    {
                        "first": "Sebastian",
                        "middle": [],
                        "last": "Jack W Rae",
                        "suffix": ""
                    },
                    {
                        "first": "Trevor",
                        "middle": [],
                        "last": "Borgeaud",
                        "suffix": ""
                    },
                    {
                        "first": "Katie",
                        "middle": [],
                        "last": "Cai",
                        "suffix": ""
                    },
                    {
                        "first": "Jordan",
                        "middle": [],
                        "last": "Millican",
                        "suffix": ""
                    },
                    {
                        "first": "Francis",
                        "middle": [],
                        "last": "Hoffmann",
                        "suffix": ""
                    },
                    {
                        "first": "John",
                        "middle": [],
                        "last": "Song",
                        "suffix": ""
                    },
                    {
                        "first": "Sarah",
                        "middle": [],
                        "last": "Aslanides",
                        "suffix": ""
                    },
                    {
                        "first": "Roman",
                        "middle": [],
                        "last": "Henderson",
                        "suffix": ""
                    },
                    {
                        "first": "Susannah",
                        "middle": [],
                        "last": "Ring",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Young",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2112.11446"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Jack W Rae, Sebastian Borgeaud, Trevor Cai, Katie Millican, Jordan Hoffmann, Francis Song, John Aslanides, Sarah Henderson, Roman Ring, Susan- nah Young, et al. 2021. Scaling language models: Methods, analysis & insights from training gopher. arXiv preprint arXiv:2112.11446.",
                "links": null
            },
            "BIBREF46": {
                "ref_id": "b46",
                "title": "Zero: memory optimization towards training a trillion parameter models",
                "authors": [
                    {
                        "first": "",
                        "middle": [],
                        "last": "S Rajbhandari",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Rasley",
                        "suffix": ""
                    },
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Ruwase",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "He",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S Rajbhandari, J Rasley, O Ruwase, and Y He. 2019. Zero: memory optimization towards training a trillion parameter models. arxiv e-prints arxiv: 11910.02054 (2019).",
                "links": null
            },
            "BIBREF47": {
                "ref_id": "b47",
                "title": "Effect of scale on catastrophic forgetting in neural networks",
                "authors": [
                    {
                        "first": "Aitor",
                        "middle": [],
                        "last": "Vinay Venkatesh Ramasesh",
                        "suffix": ""
                    },
                    {
                        "first": "Ethan",
                        "middle": [],
                        "last": "Lewkowycz",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Dyer",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "International Conference on Learning Representations",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Vinay Venkatesh Ramasesh, Aitor Lewkowycz, and Ethan Dyer. 2021. Effect of scale on catastrophic forgetting in neural networks. In International Con- ference on Learning Representations.",
                "links": null
            },
            "BIBREF48": {
                "ref_id": "b48",
                "title": "Deepspeed: System optimizations enable training deep learning models with over 100 billion parameters",
                "authors": [
                    {
                        "first": "Jeff",
                        "middle": [],
                        "last": "Rasley",
                        "suffix": ""
                    },
                    {
                        "first": "Samyam",
                        "middle": [],
                        "last": "Rajbhandari",
                        "suffix": ""
                    },
                    {
                        "first": "Olatunji",
                        "middle": [],
                        "last": "Ruwase",
                        "suffix": ""
                    },
                    {
                        "first": "Yuxiong",
                        "middle": [],
                        "last": "He",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining",
                "volume": "",
                "issue": "",
                "pages": "3505--3506",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jeff Rasley, Samyam Rajbhandari, Olatunji Ruwase, and Yuxiong He. 2020. Deepspeed: System optimiza- tions enable training deep learning models with over 100 billion parameters. In Proceedings of the 26th ACM SIGKDD International Conference on Knowl- edge Discovery & Data Mining, pages 3505-3506.",
                "links": null
            },
            "BIBREF49": {
                "ref_id": "b49",
                "title": "Biomegatron: Larger biomedical domain language model",
                "authors": [
                    {
                        "first": "Hoo-Chang",
                        "middle": [],
                        "last": "Shin",
                        "suffix": ""
                    },
                    {
                        "first": "Yang",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "Evelina",
                        "middle": [],
                        "last": "Bakhturina",
                        "suffix": ""
                    },
                    {
                        "first": "Raul",
                        "middle": [],
                        "last": "Puri",
                        "suffix": ""
                    },
                    {
                        "first": "Mostofa",
                        "middle": [],
                        "last": "Patwary",
                        "suffix": ""
                    },
                    {
                        "first": "Mohammad",
                        "middle": [],
                        "last": "Shoeybi",
                        "suffix": ""
                    },
                    {
                        "first": "Raghav",
                        "middle": [],
                        "last": "Mani",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2010.06060"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Hoo-Chang Shin, Yang Zhang, Evelina Bakhturina, Raul Puri, Mostofa Patwary, Mohammad Shoeybi, and Raghav Mani. 2020. Biomegatron: Larger biomedical domain language model. arXiv preprint arXiv:2010.06060.",
                "links": null
            },
            "BIBREF50": {
                "ref_id": "b50",
                "title": "Megatron-lm: Training multi-billion parameter language models using model parallelism",
                "authors": [
                    {
                        "first": "Mohammad",
                        "middle": [],
                        "last": "Shoeybi",
                        "suffix": ""
                    },
                    {
                        "first": "Mostofa",
                        "middle": [],
                        "last": "Patwary",
                        "suffix": ""
                    },
                    {
                        "first": "Raul",
                        "middle": [],
                        "last": "Puri",
                        "suffix": ""
                    },
                    {
                        "first": "Patrick",
                        "middle": [],
                        "last": "Legresley",
                        "suffix": ""
                    },
                    {
                        "first": "Jared",
                        "middle": [],
                        "last": "Casper",
                        "suffix": ""
                    },
                    {
                        "first": "Bryan",
                        "middle": [],
                        "last": "Catanzaro",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1909.08053"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper, and Bryan Catan- zaro. 2019. Megatron-lm: Training multi-billion parameter language models using model parallelism. arXiv preprint arXiv:1909.08053.",
                "links": null
            },
            "BIBREF51": {
                "ref_id": "b51",
                "title": "Flava: A foundational language and vision alignment model",
                "authors": [
                    {
                        "first": "Amanpreet",
                        "middle": [],
                        "last": "Singh",
                        "suffix": ""
                    },
                    {
                        "first": "Ronghang",
                        "middle": [],
                        "last": "Hu",
                        "suffix": ""
                    },
                    {
                        "first": "Vedanuj",
                        "middle": [],
                        "last": "Goswami",
                        "suffix": ""
                    },
                    {
                        "first": "Guillaume",
                        "middle": [],
                        "last": "Couairon",
                        "suffix": ""
                    },
                    {
                        "first": "Wojciech",
                        "middle": [],
                        "last": "Galuba",
                        "suffix": ""
                    },
                    {
                        "first": "Marcus",
                        "middle": [],
                        "last": "Rohrbach",
                        "suffix": ""
                    },
                    {
                        "first": "Douwe",
                        "middle": [],
                        "last": "Kiela",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2112.04482"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela. 2021. Flava: A founda- tional language and vision alignment model. arXiv preprint arXiv:2112.04482.",
                "links": null
            },
            "BIBREF52": {
                "ref_id": "b52",
                "title": "Using deepspeed and megatron to train megatron-turing nlg 530b, a large-scale generative language model",
                "authors": [
                    {
                        "first": "Shaden",
                        "middle": [],
                        "last": "Smith",
                        "suffix": ""
                    },
                    {
                        "first": "Mostofa",
                        "middle": [],
                        "last": "Patwary",
                        "suffix": ""
                    },
                    {
                        "first": "Brandon",
                        "middle": [],
                        "last": "Norick",
                        "suffix": ""
                    },
                    {
                        "first": "Patrick",
                        "middle": [],
                        "last": "Legresley",
                        "suffix": ""
                    },
                    {
                        "first": "Samyam",
                        "middle": [],
                        "last": "Rajbhandari",
                        "suffix": ""
                    },
                    {
                        "first": "Jared",
                        "middle": [],
                        "last": "Casper",
                        "suffix": ""
                    },
                    {
                        "first": "Zhun",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "Shrimai",
                        "middle": [],
                        "last": "Prabhumoye",
                        "suffix": ""
                    },
                    {
                        "first": "George",
                        "middle": [],
                        "last": "Zerveas",
                        "suffix": ""
                    },
                    {
                        "first": "Vijay",
                        "middle": [],
                        "last": "Korthikanti",
                        "suffix": ""
                    }
                ],
                "year": 2022,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2201.11990"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Shaden Smith, Mostofa Patwary, Brandon Norick, Patrick LeGresley, Samyam Rajbhandari, Jared Casper, Zhun Liu, Shrimai Prabhumoye, George Zerveas, Vijay Korthikanti, et al. 2022. Using deep- speed and megatron to train megatron-turing nlg 530b, a large-scale generative language model. arXiv preprint arXiv:2201.11990.",
                "links": null
            },
            "BIBREF53": {
                "ref_id": "b53",
                "title": "Roformer: Enhanced transformer with rotary position embedding",
                "authors": [
                    {
                        "first": "Jianlin",
                        "middle": [],
                        "last": "Su",
                        "suffix": ""
                    },
                    {
                        "first": "Yu",
                        "middle": [],
                        "last": "Lu",
                        "suffix": ""
                    },
                    {
                        "first": "Shengfeng",
                        "middle": [],
                        "last": "Pan",
                        "suffix": ""
                    },
                    {
                        "first": "Bo",
                        "middle": [],
                        "last": "Wen",
                        "suffix": ""
                    },
                    {
                        "first": "Yunfeng",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2104.09864"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Jianlin Su, Yu Lu, Shengfeng Pan, Bo Wen, and Yun- feng Liu. 2021. Roformer: Enhanced transformer with rotary position embedding. arXiv preprint arXiv:2104.09864.",
                "links": null
            },
            "BIBREF54": {
                "ref_id": "b54",
                "title": "Microsoft academic graph: When experts are not enough",
                "authors": [
                    {
                        "first": "Kuansan",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "Zhihong",
                        "middle": [],
                        "last": "Shen",
                        "suffix": ""
                    },
                    {
                        "first": "Chiyuan",
                        "middle": [],
                        "last": "Huang",
                        "suffix": ""
                    },
                    {
                        "first": "Chieh-Han",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "Yuxiao",
                        "middle": [],
                        "last": "Dong",
                        "suffix": ""
                    },
                    {
                        "first": "Anshul",
                        "middle": [],
                        "last": "Kanakia",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Quantitative Science Studies",
                "volume": "1",
                "issue": "1",
                "pages": "396--413",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kuansan Wang, Zhihong Shen, Chiyuan Huang, Chieh- Han Wu, Yuxiao Dong, and Anshul Kanakia. 2020a. Microsoft academic graph: When experts are not enough. Quantitative Science Studies, 1(1):396-413.",
                "links": null
            },
            "BIBREF55": {
                "ref_id": "b55",
                "title": "CORD-19: The COVID-19 open research dataset",
                "authors": [
                    {
                        "first": "Lucy",
                        "middle": [
                            "Lu"
                        ],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "Kyle",
                        "middle": [],
                        "last": "Lo",
                        "suffix": ""
                    },
                    {
                        "first": "Yoganand",
                        "middle": [],
                        "last": "Chandrasekhar",
                        "suffix": ""
                    },
                    {
                        "first": "Russell",
                        "middle": [],
                        "last": "Reas",
                        "suffix": ""
                    },
                    {
                        "first": "Jiangjiang",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "Doug",
                        "middle": [],
                        "last": "Burdick",
                        "suffix": ""
                    },
                    {
                        "first": "Darrin",
                        "middle": [],
                        "last": "Eide",
                        "suffix": ""
                    },
                    {
                        "first": "Kathryn",
                        "middle": [],
                        "last": "Funk",
                        "suffix": ""
                    },
                    {
                        "first": "Yannis",
                        "middle": [],
                        "last": "Katsis",
                        "suffix": ""
                    },
                    {
                        "first": "Rodney",
                        "middle": [
                            "Michael"
                        ],
                        "last": "Kinney",
                        "suffix": ""
                    },
                    {
                        "first": "Yunyao",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    },
                    {
                        "first": "Ziyang",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "William",
                        "middle": [],
                        "last": "Merrill",
                        "suffix": ""
                    },
                    {
                        "first": "Paul",
                        "middle": [],
                        "last": "Mooney",
                        "suffix": ""
                    },
                    {
                        "first": "Dewey",
                        "middle": [
                            "A"
                        ],
                        "last": "Murdick",
                        "suffix": ""
                    },
                    {
                        "first": "Devvret",
                        "middle": [],
                        "last": "Rishi",
                        "suffix": ""
                    },
                    {
                        "first": "Jerry",
                        "middle": [],
                        "last": "Sheehan",
                        "suffix": ""
                    },
                    {
                        "first": "Zhihong",
                        "middle": [],
                        "last": "Shen",
                        "suffix": ""
                    },
                    {
                        "first": "Brandon",
                        "middle": [],
                        "last": "Stilson",
                        "suffix": ""
                    },
                    {
                        "first": "Alex",
                        "middle": [
                            "D"
                        ],
                        "last": "Wade",
                        "suffix": ""
                    },
                    {
                        "first": "Kuansan",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Proceedings of the 1st Workshop on NLP for COVID-19 at ACL 2020, Online",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lucy Lu Wang, Kyle Lo, Yoganand Chandrasekhar, Russell Reas, Jiangjiang Yang, Doug Burdick, Darrin Eide, Kathryn Funk, Yannis Katsis, Rodney Michael Kinney, Yunyao Li, Ziyang Liu, William Merrill, Paul Mooney, Dewey A. Murdick, Devvret Rishi, Jerry Sheehan, Zhihong Shen, Brandon Stilson, Alex D. Wade, Kuansan Wang, Nancy Xin Ru Wang, Christopher Wilhelm, Boya Xie, Douglas M. Ray- mond, Daniel S. Weld, Oren Etzioni, and Sebastian Kohlmeier. 2020b. CORD-19: The COVID-19 open research dataset. In Proceedings of the 1st Work- shop on NLP for COVID-19 at ACL 2020, Online. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF56": {
                "ref_id": "b56",
                "title": "Crowdsourcing multiple choice science questions",
                "authors": [
                    {
                        "first": "Johannes",
                        "middle": [],
                        "last": "Welbl",
                        "suffix": ""
                    },
                    {
                        "first": "F",
                        "middle": [],
                        "last": "Nelson",
                        "suffix": ""
                    },
                    {
                        "first": "Matt",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Gardner",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1707.06209"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Johannes Welbl, Nelson F Liu, and Matt Gardner. 2017. Crowdsourcing multiple choice science questions. arXiv preprint arXiv:1707.06209.",
                "links": null
            },
            "BIBREF57": {
                "ref_id": "b57",
                "title": "Songfang Huang, and Fei Huang. 2021. Improving biomedical pretrained language models with knowledge",
                "authors": [
                    {
                        "first": "Zheng",
                        "middle": [],
                        "last": "Yuan",
                        "suffix": ""
                    },
                    {
                        "first": "Yijia",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "Chuanqi",
                        "middle": [],
                        "last": "Tan",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2104.10344"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Zheng Yuan, Yijia Liu, Chuanqi Tan, Songfang Huang, and Fei Huang. 2021. Improving biomedical pre- trained language models with knowledge. arXiv preprint arXiv:2104.10344.",
                "links": null
            },
            "BIBREF58": {
                "ref_id": "b58",
                "title": "Aligning books and movies: Towards story-like visual explanations by watching movies and reading books",
                "authors": [
                    {
                        "first": "Yukun",
                        "middle": [],
                        "last": "Zhu",
                        "suffix": ""
                    },
                    {
                        "first": "Ryan",
                        "middle": [],
                        "last": "Kiros",
                        "suffix": ""
                    },
                    {
                        "first": "Rich",
                        "middle": [],
                        "last": "Zemel",
                        "suffix": ""
                    },
                    {
                        "first": "Ruslan",
                        "middle": [],
                        "last": "Salakhutdinov",
                        "suffix": ""
                    },
                    {
                        "first": "Raquel",
                        "middle": [],
                        "last": "Urtasun",
                        "suffix": ""
                    },
                    {
                        "first": "Antonio",
                        "middle": [],
                        "last": "Torralba",
                        "suffix": ""
                    },
                    {
                        "first": "Sanja",
                        "middle": [],
                        "last": "Fidler",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Proceedings of the IEEE international conference on computer vision",
                "volume": "",
                "issue": "",
                "pages": "19--27",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yukun Zhu, Ryan Kiros, Rich Zemel, Ruslan Salakhut- dinov, Raquel Urtasun, Antonio Torralba, and Sanja Fidler. 2015. Aligning books and movies: Towards story-like visual explanations by watching movies and reading books. In Proceedings of the IEEE in- ternational conference on computer vision, pages 19-27.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "num": null,
                "type_str": "figure",
                "text": "Sources with <1% (upper right, black box) Figure 1: Summary of data source representation within the Combined A+F data sample. Coloring illustrates whether a data source contains peer reviewed (Blue), mixed (Purple), or not peer reviewed (Red) articles.",
                "uris": null
            },
            "FIGREF1": {
                "num": null,
                "type_str": "figure",
                "text": "Distribution of validation loss by model size: performance improves as the model size increases.",
                "uris": null
            },
            "FIGREF3": {
                "num": null,
                "type_str": "figure",
                "text": "(a) GPU computation in #Floating Point Operations (b) GPU Memory Allocation Figure 4: GPU system performance during pretraining.",
                "uris": null
            },
            "TABREF0": {
                "html": null,
                "type_str": "table",
                "content": "<table><tr><td>Model</td><td/><td>Data Source</td><td>Pretraining</td><td>Corpus</td><td>#Params (B)</td></tr><tr><td colspan=\"5\">Lee et al. 2020 Alsentzer et al. 2019 Peng et al. 2019 Liu et al. 2021 Guo et al. 2021 Chem(Rxn)BERT  \u2020 Wiki + Books continual pretraining Chemistry Journals BioBERT Wiki + Books continual pretraining PubMed ClinicalBERT Wiki + Books continual pretraining MIMIC 1 BlueBERT Wiki + Books continual pretraining PubMed + MIMIC MATH-BERT Arxiv continual pretraining Arxiv Phan et al. 2021 SciFive C4 continual pretraining PubMed Naseem et al. 2021 BioALBERT Wiki + Books continual pretraining PMC + MIMIC-II Lewis et al. 2020 BioRoBERTa Wiki + Books continual pretraining PMC + MIMIC-III Yuan et al. 2021 KeBioLM PubMed continual pretraining PubMed + UMLS 2</td><td>0.11 0.11 0.11 0.11 0.11 0.22 0.77 0.02 0.30 0.34</td></tr><tr><td>Shin et al. 2020 Kanakarajan et al. 2021 Miolo et al. 2021 Beltagy et al. 2019 Liu et al. 2021 Gu et al. 2021</td><td>BioMegatron BioELECTRA ELECTRAMed SciBERT OAG-BERT PubMedBERT</td><td>PubMed PubMed PubMed PMC + CS OAG PubMed</td><td>from scratch continual pretraining from scratch from scratch from scratch from scratch from scratch</td><td>PubMed PubMed PubMed PMC + CS OAG PubMed</td><td>0.80 1.20 0.11 0.11 0.11 0.11 0.34</td></tr><tr><td colspan=\"2\">Our Work (autoregressive)  \u2020</td><td>10+ sources (Chemistry)</td><td>from scratch continual pretraining</td><td>10+ sources (Chemistry)</td><td>1.47</td></tr></table>",
                "text": "Foundation models for science focus on the biomedical, math, computer science and chemistry domains. We use \u2020 to indicate models trained for chemistry.",
                "num": null
            },
            "TABREF1": {
                "html": null,
                "type_str": "table",
                "content": "<table><tr><td colspan=\"2\">We compare model configurations between GPT-NeoX and OpenAI's GPT-2. GPT-NeoX architecture is originally from GPT-3 (Brown et al., 2020)</td></tr><tr><td>Size S M L XL</td><td>Model GPT-NeoX 12 dL d dim d heads #Params (B) 768 12 0.18 GPT-2 12 768 12 GPT-NeoX 24 1024 16 0.40 GPT-2 24 1024 16 GPT-NeoX 24 1536 16 0.80 GPT-2 36 1280 20 GPT-NeoX 24 2048 16 1.47 GPT-2 48 1600 25</td></tr></table>",
                "text": "Our model configurations: d L is the number of decoder layers, d dim is the hidden size of the model, d heads is the number of attention heads.",
                "num": null
            },
            "TABREF2": {
                "html": null,
                "type_str": "table",
                "content": "<table><tr><td>Source MAG \u2329A\u232a Aminer \u2329A\u232a S2ORC \u2329A\u232a WoS \u2329A\u232a CORD-19 \u2329A\u232a OSTI \u2329A\u232a Arxiv \u2329A\u232a PubMed \u2329A\u232a PubMed \u2329FT\u232a CORE \u2329FT\u232a Combined \u2329A\u232a Combined \u2329FT\u232a Combined \u2329A+FT\u232a</td><td>#Articles (M) #Tokens (B) Size (Gb) 34.26 7.43 46 18.50 5.80 35 10.44 2.05 32 7.90 3.31 18 &lt; 0.01 &lt; 0.01 0.2 0.05 &lt; 0.01 0.1 0.38 0.04 0.4 0.28 0.08 0.5 0.70 7.34 32 7.27 215.50 743 46.94 16.18 67 6.52 184.42 603 53.45 200.61 670</td></tr></table>",
                "text": "Dataset statistics: combined datasets are after the de-duplication process. We split datasets to those that include abstracts \u2329A\u232a vs. full texts \u2329FT\u232a.",
                "num": null
            },
            "TABREF4": {
                "html": null,
                "type_str": "table",
                "content": "<table><tr><td>Model</td><td>Size</td><td>BoolQ</td><td colspan=\"6\">CB WIC WSC MathQA PIQA PubMedQA</td><td>Lambada Wikitext</td></tr><tr><td colspan=\"2\">Baseline AMiner CORE MAG PubMed-F S2ORC WoS Combined-A Combined-F Combined-A+F XL S M L XL M \u2021 S M L XL S M L XL S M L XL S M L XL S M L XL S M L XL XL XL Combined-A+F XL (4x)</td><td colspan=\"2\">0.49 0.41 0.59 0.43 0.60 0.45 0.61 0.39 0.62 0.34 0.41 0.39 0.40 0.39 0.61 0.48 0.50 0.39 0.62 0.41 0.62 0.41 0.61 0.41 0.61 0.38 0.41 0.23 0.38 0.07 0.51 0.14 0.40 0.11 0.58 0.41 0.61 0.39 0.57 0.41 0.60 0.41 0.38 0.41 0.38 0.43 0.38 0.46 0.38 0.50 0.38 0.39 0.38 0.45 0.41 0.36 0.57 0.34 0.56 0.16 0.62 0.38 0.61 0.41 0.61 0.41</td><td>0.49 0.50 0.50 0.50 0.50 0.50 0.51 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.51 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.47 0.50 0.50 0.50 0.50 0.50</td><td>0.43 0.40 0.46 0.50 0.36 0.44 0.41 0.47 0.37 0.37 0.37 0.37 0.37 0.40 0.37 0.35 0.62 0.45 0.38 0.38 0.39 0.63 0.63 0.63 0.63 0.63 0.63 0.54 0.37 0.37 0.37 0.39 0.37</td><td>0.21 0.23 0.23 0.24 0.20 0.22 0.21 0.22 0.21 0.20 0.21 0.21 0.22 0.21 0.21 0.22 0.22 0.21 0.20 0.21 0.22 0.20 0.22 0.21 0.20 0.21 0.19 0.21 0.20 0.21 0.22 0.23 0.24</td><td>0.63 0.68 0.70 0.71 0.55 0.56 0.57 0.58 0.58 0.55 0.56 0.57 0.58 0.56 0.57 0.59 0.59 0.57 0.58 0.59 0.59 0.57 0.56 0.56 0.56 0.55 0.54 0.56 0.55 0.60 0.57 0.59 0.60</td><td>0.44 0.53 0.54 0.59 0.55 0.46 0.43 0.36 0.43 0.55 0.55 0.51 0.45 0.43 0.41 0.39 0.34 0.54 0.49 0.42 0.49 0.34 0.34 0.34 0.33 0.34 0.34 0.42 0.56 0.50 0.55 0.48 0.56</td><td>40.06 18.25 12.97 10.63 2834.51 2825.84 1802.35 661.81 786.22 671.43 273.06 173.15 79.95 1142.83 628.72 282.39 364.54 2670.39 1742.00 843.83 679.80 122739.30 80151.10 89136.68 107065.48 140552.69 182967.37 148609.73 192970.64 250.88 72.50 71.43 30.40</td><td>37.37 26.75 22.61 20.38 126.55 158.85 116.93 87.23 91.28 100.53 77.96 69.62 50.47 118.40 91.36 67.74 70.71 148.88 119.74 95.75 90.38 403.48 330.56 327.53 351.81 556.00 498.36 480.91 509.06 61.07 48.96 48.65 33.05</td></tr></table>",
                "text": "Downstream Out-of-domain Task Performance. We use \u2021 to indicate the baseline model tuned from the base GPT-2 model. Performance on Lambada and Wikitext is reported using perplexity, all other tasks report accuracy . Top-4 performance highlighted in bold, with best performance indicated with underlines. XL (4x) model is trained with 4x larger batch size that used in other models.",
                "num": null
            }
        }
    }
}