Add chat template to tokenizer config
#2
by
lewtun
HF staff
- opened
Great work with Tulu 2 folks!
FYI it would be great to include the chat template you used to train the model directly in the tokenizer_config.json so that it's easier to run inference. See the docs here: https://huggingface.co/docs/transformers/v4.35.2/en/chat_templating#advanced-how-do-chat-templates-work
You can also see how we added the template for Zephyr 7B here: https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/blob/c2339b671588a96001b9766328dd2ef4b82ef5c4/tokenizer_config.json#L34
Thanks for the tip! Added :)
hamishivi
changed discussion status to
closed