Add chat templates to tokenizer_config.json

#46
by Rocketknight1 HF staff - opened

Hi, Matt from Hugging Face here! We held off on doing this for a while, since multiple templates weren't properly supported, but they're in the 4.39 release that's been out for a couple of weeks, so it might be safe to start changing over.

This PR just copies the default chat templates from the tokenizer class into tokenizer_config.json, making them accessible to users and to conversion tools (e.g. llama.cpp). Over time, we hope to deprecate hardcoded templates entirely!

Cohere For AI org

Thanks a lot Matt, this makes sense and looks very nice. I'm merging your PR.

ahmetustun changed pull request status to merged
Cohere For AI org
edited Apr 4

Hey Matt! After I merged the changes, automatic parsing gave this error:
Screenshot 2024-04-04 at 20.16.38.png

Sign up or log in to comment