NVIDIA-NeMo / NVIDIA-NeMo/RL

token batch size rather than GBS

Open
#332 2 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
2k
Forks
561
Avg merge
4d 5h
Merged PRs (30d)
145

Description

if the generation has high variance, our training batch size(in tokens) also has high variance.

we should define a token batch size, which our code tries it's best to achieve, i think this might stabilize training

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.