Although scaling Transformer-based generators has been central to recent advances, the tokenizer component itself is rarely scaled, leaving open questions about how auto-encoder design choices ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results