The first Dataset in the training reset to the first on the dataset list when editing the configuration.

[alcaitiff]

This is for bugs only

Did you already ask in the discord?

Yes

You verified that this is a bug and not a feature request or question by asking in the discord?

Yes

Describe the bug

Steps to reproduce:

I have five datasets available in my workspace, listed in alphabetical order: A, B, C, D, and E.
I created a LoRA training job using datasets C and E (selected in that specific order: first C, then E).
After the job was done or paused, the job details page correctly shows the datasets used as C and E, in the expected order.

Issue:
When I click the "Edit" (pencil) button to modify the job configuration:

The first dataset in the configuration is sometimes automatically changed from C to A (the first dataset in the overall list of available datasets).
The second dataset remains E.
As a result, the edited configuration shows datasets A and E instead of the original C and E.

From my experience, this behavior occurs very often but not every single time (it's intermittent). This makes it harder to reproduce consistently, which I understand could complicate debugging and fixing the issue.
This appears to override the originally selected dataset with the first one in the alphabetical list in many cases when the job is opened for editing.
Could you please investigate this when possible? It affects the reliability of editing jobs .

[ZeTofZone]

Hi.
I can confirm that I have the same behavior. This is random, and I can't define exactly when it happens, but I took the use to verify each time I'm pausing a job if the first selected dataset was the one I wanted before resuming it.

[mcDandy]

It can happen when you paste a training config to advanced (or only relevant part) or when clonning a job.

[thangquay347]

This used to happen to me using an older version of ai toolkit, updated to the newest version managed to get rid of this issue.