Docs: Megatron tutorial references outdated repository structure

[kimmeoungjun]

Is there a policy or guideline for keeping DeepSpeed tutorials up to date?

In docs/_tutorials/megatron.md, the tutorial references files such as arguments.py, pretrain_gpt2.py, and utils.py from the deepspeedai/Megatron-DeepSpeed repository. However, these references no longer exactly match the current file structure of the Megatron-DeepSpeed repository.

While it is possible to map the tutorial’s descriptions to the current codebase with some investigation, it is natural for small mismatches to accumulate as the project evolves. Ideally, a tutorial should be directly runnable and aligned with the current state of the project.

My question is: when such inconsistencies are found, is it acceptable to update the tutorial incrementally as needed? If there is an existing policy or preferred approach for handling these documentation updates, I would appreciate any guidance.