Create a separate `eval` function to replace `train` + `eval_mode=True`
Open
- Dominant language
- Python
- Stars
- 2k
- Forks
- 561
- Avg merge
- 4d 5h
- Merged PRs (30d)
- 145
Description
See https://github.com/NVIDIA/NeMo-RL/pull/446#pullrequestreview-2872069062 -- when running interleaved evaluation during training, we use the same `train` function with `eval_mode=True`, which is starting to become messy. We should break the eval case out into a separate function.
Contributor guide
Assessment
This issue has not been assessed yet.