Resumable experiments?

[andstor]

Is there any way to resume an experiment? For long-running experiments, it would be very useful to be able to resume an interrupted experiment run.

[dosubot[bot]]

Phoenix doesn't have explicit support for resumable experiments or checkpointing. However, it does persist each completed run to the backend as it finishes. If an experiment is interrupted, you can safely rerun it with the same parameters—Phoenix will skip already-completed runs and only process the missing ones, thanks to API deduplication and conflict handling. This provides a form of implicit resumability for long-running experiments, but there's no formal API for pausing or resuming at a specific point, nor granular control over experiment state. For more advanced checkpointing or orchestration, you'd need to implement external logic or use a workflow manager. You can always retrieve experiment state and results by experiment ID for further evaluation or analysis.
Source: Phoenix experiment code and docs

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[axiomofjoy]

Makes a lot of sense @andstor especially if you're running an experiment on a large dataset. We have gotten a similar ask here.

Can you add a comment to that issue describing your needs in more detail? In particular, we'd be interested in knowing whether you're running your experiments via the client or via the playground UI, or both.