That makes sense, thanks for the detailed reply. I’m curious about the latest research related to continuously improving the tokenizer while training the main

That makes sense, thanks for the detailed reply. I’m curious about the latest research related to continuously improving the tokenizer while training the main model (or something equivalent).