Chat/instruction template

#2
by Gnurro2 - opened

Which (multiturn) chat or instruction format was used to train/finetune the model? The base model being Llama2 suggests the Llama2 template, but the datasets used for finetuning are associated with other prompt formats...

tokyotech-llm org
edited May 3

@Gnurro2
Hi,
We have recently released a better instruction model:
https://huggingface.co/tokyotech-llm/Swallow-70b-instruct-v0.1

We recommend using this model.
The latest model excels in responsiveness to multi-turn instructions and more. For more details, please visit:

https://huggingface.co/tokyotech-llm/Swallow-70b-instruct-v0.1#instruction-format-ver01"

As a note, the format is different from that of Llama2.

Sign up or log in to comment