Implement NMF based on attention layer in transformers
Implement NMF based on attention layer in transformers