An implementation of local windowed attention for language modeling - View it on GitHub