This model was trained using the LOGION tokenizer, but the last ids of it were used to indicate masks (that is, the first mask is id 50_000, the second mask is 49_999, the third mas is 49_998 etc.).
This model was trained using the LOGION tokenizer, but the last ids of it were used to indicate masks (that is, the first mask is id 50_000, the second mask is 49_999, the third mas is 49_998 etc.).