compute relative position buckets for relative attention bias

[pmabbo13]

Stack from ghstack (oldest at bottom):

• add t5 model that can function as both encodery-only or encoder-decoder model #1829
• add t5 stack that can function as either the encoder or decoder of a t5 model #1828
• add t5 layer module that can be used for both encoder or decoder stack #1827
• add layer norm module for t5 model #1826
• adding forward method for multihead attention #1833
• computing attention scores using relative attention bias #1832
• computing relative attention bias #1831
• -> compute relative position buckets for relative attention bias #1830
• create T5MultiheadAttention module #1825

[pmabbo13]

Description

This method creates the assigned buckets to which ranges of relative positions between tokens will be assigned. The implementation is taken directly from HF

[parmeet]

LGTM!