Byte Latent Transformer: Patches Scale Better Than Tokens(ai.meta.com)
366 points by zxexz | 83 comments
Meta AI research publication page for the Byte Latent Transformer paper, which proposes byte-level patches instead of tokens for scaling language models.