openai/tiktoken ? reverse-engineered prompt

I need a Python library that can quickly count tokens for OpenAI's models. It should be optimized for performance, much faster than other common tokenizers. The core feature should be a way to get the correct tokenizer just by specifying the model name, like "gpt-4o" or "gpt-4".

Once I have the tokenizer, I want to be able to do two things: encode any string of text into its corresponding list of token IDs, and then decode that list of token IDs back into the original text. This is super important for prepping data for the API and making sure my inputs fit inside the context window.

Please build this as a standard Python package that I can install directly with pip. It should be a simple, efficient tool for working with OpenAI's tokenization scheme.