Enable rpe methods in bert-like models

[sararb]

What does this PR do ?

• Add extrapolatable position embedding to bert-like model, based on the recent changes introduced in GPT extrapolatable position embedding (xpos/sandwich/alibi/kerple) and Flash Attention #6666
• I am already using this change in the BioNeMo framework and running experiments to test how rpe techniques help bert-like model to extrapolate to longer sequences.

Collection: [NLP]

Changelog

• BERT:
  • nemo/collections/nlp/models/language_modeling/megatron_bert_model.py
  • nemo/collections/nlp/models/language_modeling/megatron/bert_model.py

Usage

model.position_embedding_type='' # Options ['learned_absolute', 'rope', 'alibi', 'xpos', 'sandwich', 'kerple']

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

• Related to # (issue)