NVIDIA / NVIDIA/Megatron-LM

Improve Megatron-RL testing

Open
#2,460 0 comments 0 reactions 4 assignees Claimed by @tdene View on GitHub
enhancement module: rl
Dominant language
Python
Stars
17.9k
Forks
4.5k
Avg merge
4d 3h
Merged PRs (30d)
272

Description

The native RL training loop in Megatron-RL is vulnerable to breaking changes due to insufficient test coverage. This issue is for tracking tests we wish to add:

- [ ] unit test to trigger alert when default params for RL training are changed (in args or hardcoded, e.g. in inference)
- [ ] functional test with high memory pressure model to alert to changes in memory consumption
- [ ] add functional tests for the sequence parallelism plus sequence packing case

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.