Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
They follow the architecture described in Generating Long Sequences with Sparse Transformers, modified to support longer context length.