For llama2 and llama 3, add the running configuration file and add th…

FlagOpen / FlagScale

FlagScale is a large model toolkit based on open-sourced projects.

Other

132 stars 40 forks source link

Closed phoenixdong closed 4 months ago

phoenixdong commented 4 months ago

Added yaml configuration file that can train Llama2.
A conversion script has been added that can be used to convert ckpt in Hugging Face format and ckpt in mcore format to and from each other.
Added yaml configuration file that can finetune Llama3.