NVIDIA-NeMo / NVIDIA-NeMo/RL

gpt-oss-20 OOMs on 8 B200s using default config

Open
#2,157 1 comment 0 reactions 1 assignee Claimed by @RayenTian View on GitHub
bug community-request waiting-on-maintainers
Dominant language
Python
Stars
2k
Forks
562
Avg merge
4d 1h
Merged PRs (30d)
150

Description

Hi!

I'm using the provided config values for gpt-oss but despite TP+EP across 8 B200s (and using flex attention) I keep running into OOMs; I've also tried configuring megatron as well but that just seems to outright fail as it falls back to a naive attention implementation.

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.