Len(trainer.model.state_dict().keys()) reduced after calling trainer.train()

Hello everyone,

I am trying to finetune a LLaMA based model on my local server.

The issue I have is that after calling trainer.train(), the number of keys in trainer.model.state_dict() reduced. I noticed that all the missing keys are end with .bias.

I would appreciate it if you guys could suggest any solutions.

Thank you.