I’m writing a light-weight javascript library that could be used by other LLM’s as well. I don’t want to force a dependency on a specific tokenizer. The library user can specify the tokenizer, but I want to provide a simple default that will be in the right ballpark.