After wrapped by FSDP, most of my model's weights are empty using named_parameters()

I have a LLM and trained it using transformers with FSDP, I found that after being wrapped by FSDP, most of my model’s weights are empty using named_parameters(), see figure below:


Could you tell me what is the correct way to get all model {name: param} dict after being wrapped by FSDP, thanks a lot