Nano v3 dpo failed with cpu-offload after transformer v5 bump.
Open
bug
- Dominant language
- Python
- Stars
- 2k
- Forks
- 561
- Avg merge
- 4d 5h
- Merged PRs (30d)
- 145
Description
For now we doesn't use cpu-offload in `examples/configs/recipes/llm/dpo-nanov3-30B3AB-1n8g-fsdp8ep8-automodel.yaml`.
After it fixed, need to add cpu-offload in this config.
Related PR: https://github.com/NVIDIA-NeMo/RL/pull/1962
Contributor guide
Assessment
This issue has not been assessed yet.