NVIDIA / NVIDIA/Megatron-LM

Separate checkpoint type execution paths

Open
#3,758 1 comment 0 reactions 0 assignees View on GitHub
enhancement
Dominant language
Python
Stars
17.9k
Forks
4.5k
Avg merge
4d 3h
Merged PRs (30d)
272

Description

Checkpointing code is pretty unreadable because of a lot of branching in low-level functions for different checkpoint types (local, global, dist ckpt types). We should move the branching to the higher stack level, i.e. have a separate function/path for local checkpoint, global checkpoint, etc. For logic that is definitely common across paths, use helpers.

To be clear, this is not a feature request. More of a readability refactor.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.