chunked/nltk_chunking/_model_summary/chunk_47.txt

ALBERT reduces memory consumption by lowering the number of parameters in two ways: separating the larger vocabulary embedding into two smaller matrices and allowing layers to share parameters.