Distributed optimizer support for experimental FP8 tensors

[timmoon10]

What does this PR do ?

This PR integrates with Transformer Engine's experimental Float8Tensors. This allows the model to only store FP8 weight matrices. The distributed optimizer stores an FP32 master copy of the weights and performs param all-gathers in FP8.

Collection: NLP

Changelog

• Adds distributed optimizer support for FP8 parameters
• Adds the option to initialize GPT with FP8 parameters

Usage

Run GPT, e.g. with the config at https://github.com/NVIDIA/NeMo/blob/main/examples/nlp/language_modeling/conf/megatron_gpt_config.yaml.

Enable FP8 support with model.fp8=True, FP8 parameters with model.fp8_params=True, and the distributed optimizer with model.optim.name=distributed_fused_adam.

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

• Depends on Distributed optimizer support for contiguous param buffer with FP8 params apex#1749
• Uses Transformer Engine Float8Tensor added in [PyTorch] Experimental FP8 tensor class TransformerEngine#452
• Closes Distributed optimizer support for experimental FP8 tensors #7469 and Distributed optimizer support for experimental FP8 tensors (r1.20.0 branch) #7565

[github-actions]

This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.

[github-actions]

This PR was closed because it has been inactive for 7 days since being marked as stale.