This is a 800M parameters model pre-trained with QuEST over 80B C4 tokens in INT1 format (W1A1).
The code to verify that this model works in INT4 can be found here.
