Add "offline" data cache generation support

[dimapihtar]

What does this PR do ?

1. Adds "offline" data cache generation support. The training script will be stopped once data cache is generated.
2. Enable mcore's num_dataset_builder_threads param usage.

Collection: [Note which collection this PR will affect]

Changelog

• Add specific line by line info of high level changes in this PR.

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this 

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

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)

[samskillman]

What is the reasoning behind overloading num_workers for the num_dataset_builder_threads, instead of passing num_dataset_builder_threads in as a separate config parameter?

I could imagine a case where I set num_dataset_builder_threads to a very large number (64 or 128) but wouldn't want that many workers during the typical data loading. That would allow many workloads to specify a high num_dataset_builder_threads without needing to explicitly use the data_cache_generation_only offline step.

Instead I would suggest num_dataset_builder_threads as an explicit new parameter in megatron_gpt_config.yaml, and default it to 1 to mirror https://github.com/NVIDIA/Megatron-LM/blob/e33c8f78a35765d5aa37475a144da60e8a2349d1/megatron/core/datasets/blended_megatron_dataset_config.py#L48

[dimapihtar]

What is the reasoning behind overloading num_workers for the num_dataset_builder_threads, instead of passing num_dataset_builder_threads in as a separate config parameter?

I could imagine a case where I set num_dataset_builder_threads to a very large number (64 or 128) but wouldn't want that many workers during the typical data loading. That would allow many workloads to specify a high num_dataset_builder_threads without needing to explicitly use the data_cache_generation_only offline step.

Instead I would suggest num_dataset_builder_threads as an explicit new parameter in megatron_gpt_config.yaml, and default it to 1 to mirror https://github.com/NVIDIA/Megatron-LM/blob/e33c8f78a35765d5aa37475a144da60e8a2349d1/megatron/core/datasets/blended_megatron_dataset_config.py#L48

thank you for the review, fixed.

[athitten]

LGTM, thanks @dimapihtar !