guide: revisit Checkpoints guide(s?)

[jorgeorpinel]

Per task in #2548

Trying to make it broader (more general) yet shorter, and cover all the usage details (extract from cmd ref) in a structured way.

• General text/story review - guide: Review of Checkpoints guide + links #2910
• Add info from Add checkpoint backup #2672 to the checkpoints page.
• Incorporate/merge material from https://github.com/iterative/dvc.org/pull/2753/files
• Should close guide: all the ways to use checkpoints #2432
• Should close guide: Add checkpoints example that runs for a fixed number of epochs #2568
• Remove deep details from https://dvc.org/doc/command-reference/exp/run#checkpoints
• Sharing checkpoints with dvc exp push (here or in the Sharing Exps guide)
  (See guide: revisit Checkpoints guide(s?) #2769 (comment))
• Add Checkpoints sections to all other Exp guides? Per guide: revisit Checkpoints guide(s?) #2769 (comment)

---

Decide on these:

• Keep a separate guide but have some basics in a Checkpoints section in each Exps guide (Running, Sharing, etc.)
• Maybe guide: clarify resuming (checkpoints) #2742
• Maybe guide: Add checkpoints example that runs for a fixed number of epochs #2568
• Maybe guide: checkpoints for Tensorflow  #2509
• Maybe guide: Add checkpoints example with downstream stages #2741

[daavoo]

It might be worth adding a mention about how to properly Share Checkpoints with dvc exp push.

The current content finishes without sharing the checkpoints and the last section Adding checkpoints to git will squash the checkpoint commits making them not usable later.

We kind of already have the Sharing Experiments guide but it doesn't mention Checkpoints at all.

[daavoo]

As new Checkpoint specific issues appear (#2797) I think it would make more sense to have a dedicated Checkpoints section on each individual Experiment guide (Running, Sharing, etc.), like https://dvc.org/doc/user-guide/experiment-management/running-experiments#checkpoint-experiments .

[iesahin]

I think that's a good idea.

[jorgeorpinel]

more sense to have a dedicated Checkpoints section on each individual Experiment guide (Running, Sharing, etc.)

Would probably need both its own guide and a small Checkpoints section in each Exps guide.

[daavoo]

more sense to have a dedicated Checkpoints section on each individual Experiment guide (Running, Sharing, etc.)

Would probably need both its own guide and a small Checkpoints section in each Exps guide.

For the standalone checkpoints guide, I would recommend replacing the PyTorch code snippets with one of the frameworks that have a dvclive callback integration.

This would reduce the number of unrelated details that are in the current guide.

[jorgeorpinel]

Should we merge this issue with #2432?

[jorgeorpinel]

@daavoo want to repurpose this to include checkpoint charing info for now? Per #2769 (comment)
BTW exp show tables or other code samples may need updating too?

Otherwise this looks stale (see #2432 (comment) and #2568 (comment)) and could be closed (updating the doc in question will be a natural result of updating the core feature anyway).