This is a heavily interactive web application, and JavaScript is required. Simple HTML interfaces are possible, but that is not what this is.
Post
Ted Underwood
tedunderwood.com
did:plc:565ebob5f6hw33hjdkxty6qj
Subword tokenization is weird and ad-hoc. Logically, it would make sense to divide into bytes—except then, sequences would be too long for transformers. But not too long for state-space models like Mamba! 🤖 #MLSky arxiv.org/abs/2401.13660
2024-01-25T02:59:41.550Z