# Tainer.train() - save\_total\_limit - latest vs best

**URL:** </static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftainer-train-save-total-limit-latest-vs-best%2F18636%3E%5C
**Category:** Beginners\
**Created:** [June 2, 2022, 12:30am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftainer-train-save-total-limit-latest-vs-best%2F18636 "2022-06-02T00:30:10Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![rpu](https://avatars.discourse-cdn.com/v4/letter/r/5fc32e/32.png) [@rpu](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Frpu)%5C
**Post date:** [June 2, 2022, 12:30am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftainer-train-save-total-limit-latest-vs-best%2F18636%2F1 "2022-06-02T00:30:10Z")

</div>

I believe save\_total\_limit along=N with metric\_for\_best\_model=metric will result in the trainer keeping the latest N best checkpoints per the specified metric.

To continue training in case of crashes, I’d like to also keep the most recent checkpoint regardless of its performance. Is that possible without customizing the trainer?

thanks!