chunked/nltk_chunking/model_doc_fuyu/chunk_9.txt

The reason is that the model will first be downloaded ( using the dtype of the checkpoints online) then it will be cast to the default dtype of torch (becomes torch.float32).