Sampling strategies

Is it possible to set a customizable training sampler in the Trainer object?
It was possible in earlier versions of Hugging Face (4.9.0 for instance) with the train_sampler argument, but now I don’t find it anymore…
Thanks !

Actually, just read the code… and disappointed
We need to subclass and override the Trainer class just to customize a sampler (as indicated in the comments of the code)