Create causal mask on MCore/TE side

[janekl]

What does this PR do ?

Currently Megatron-LM creates causal attention mask for a local-layer spec model on its own -- see NVIDIA/Megatron-LM@a45805a -- and hence attention_mask=None can be specified in Nemo. In this MR, I'm using this new M-LM functionality when creating fwd_output_only_func function.

Broader functional tests for fwd_output_only_func will follow in subsequent MRs.

Collection: NLP

Jenkins CI

To run Jenkins, a NeMo User with write access must comment jenkins on the PR.

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 NVIDIA/Megatron-LM@a45805a

[janekl]

@titu1994 could you help to land it? Prerequisite is mcore=0.5.0 #8465

[dimapihtar]

LGTM. Thank you!