Xenova

/

gpt-4

like 38

Not a general way that I know of, but the original gpt4 tokenizer is already in tiktoken format, which you can use. Do you have a specific tokenizer in mind?

the original gpt4 tokenizer is already in tiktoken format,

Right, I was more thinking about training a HF tokenizer from scratch and creating a tiktoken model from that. Thanks 🙂

I'd love to see that, the performance gains for semantic chunking with semchunk would be great since tiktoken is much faster than transformers when it comes to tokenization.