This is a 800M parameters model pre-trained with QuEST over 80B C4 tokens in 2:4 sparse INT4 format.
The code to verify that this model works in INT4 can be found here.
