trl-lib/stack-llama · Questions Regarding the Tokenizer Used in Stack Llama

Hello,

Firstly, thank you for sharing such a valuable project.
(However, it seems that there is an error with the inference in the current space, and it is not functioning properly. There seems to be an issue with the inference in the space.)

I have some questions about stack llama:

I would like to know detailed information about the tokenizer used in stack llama. Did you use the same tokenizer as in the llama2 instruct or chat models?
What is the training data template for stack llama? Is it okay to proceed with training without using llama2-coder as the backbone?
I am interested in additional updates about stack llama.
I am preparing additional research related to DPO and would like to know whom to email for a prompt response. I wish to share a draft of my research and receive feedback. However, if it's not possible, please feel free to ignore question 4.

Thank you.