How to run the model in float32?

[xbq1994]

How to run the model in float32?

[wangjiongw]

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.