[QUESTION] Limit Number of Saved Checkpoints
Open
enhancement
- Dominant language
- Python
- Stars
- 17.9k
- Forks
- 4.5k
- Avg merge
- 4d 6h
- Merged PRs (30d)
- 271
Description
Hi,
I'm wondering if Megatron-LM supports keeping only the latest N checkpoints. When using a small `save_interval` for frequent checkpointing, the storage occupied by checkpoints grows rapidly. It would be very helpful if there was a parameter similar to `save_total_limit` in the Transformers library, which would allow us to constrain the maximum number of checkpoints saved. This would effectively reduce storage consumption.
Thanks!
Contributor guide
Assessment
This issue has not been assessed yet.