chunked/nltk_chunking/model_doc_longt5/chunk_23.txt

As a consequence, TGlobal attention introduces
a few new parameters -- global relative position biases and a layer normalization for global token's embedding.