NotImplementedError: offload_to_cpu=True and NO_SHARD is not supported yet

[mechigonft]

I am using a single GPU(A10) to run Bloom-560m model fine-tune, error, how to solve? I found similar problems in other projects, but I didn't know how to solve the problems in alpaca
Alpha-VLLM/LLaMA2-Accessory#76

[mechigonft]

When saving the model checkpoint, an error occurred, which should be a problem with fsdb. My Python version is 3.8.16
Can I solve this problem by deleting the "--fsdp "full_shard auto_wrap"

--fsdp_transformer_layer_cls_to_wrap 'BloomBlock' " from the bash script?