megatron.data.t5_dataset.make_attention_mask# megatron.data.t5_dataset.make_attention_mask(source_block, target_block)# Returns a 2-dimensional (2-D) attention mask :param source_block: 1-D array :param target_block: 1-D array