NVIDIA-NeMo / NVIDIA-NeMo/RL

Create a singular "to HF" checkpoint script

Open
#1,535 0 comments 0 reactions 0 assignees View on GitHub
enhancement
Dominant language
Python
Stars
2k
Forks
561
Avg merge
4d 5h
Merged PRs (30d)
145

Description

There are multiple checkpoint formats:
* dtensorv2 (safe tensors)
* dtensorv2 (DCP)
* mcore (DCP)

and we currently provide two checkpoint conversion scripts that users sometimes mistakenly apply (use mcore conversion script for Dtensor V2 DCP b/c the file looks the same).

Since there is enough metadata in the checkpoint directory, we should provide a single conversion entrypoint that automatically converts all checkpoint types.

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.