How to run the model in float32? #82

Activity

1. wangjiongw commented on Oct 23, 2024

   

   wangjiongw

   on Oct 23, 2024

   Collaborator

   More actions

   As the repo is build upon deepspeed framework. You can find corresponding config file in src/config, where you can config fp16 or bf16. I think setting both of them to False will run the model in float32. You can find more on deepspeed docs.