[Feature Request] Save/load internal states of pseudo-random number generators

[lbergmann1]

🚀 Feature

Hello,

I suggest to automatically save/load internal states of all pseudo-random number generators of a model (python, torch, numpy, action_space) in case of model.save()/model.load().

This is what I have done so far. I did not include RNGs of environments, because the model may be loaded with new training environments.

Motivation

I would like to save a checkpoint of my current model so that it is possible to continue training in a reproducible way, i.e. with the same results as if the training had never been interrupted.
Since internal states of RNGs are not saved, it is not possible to reproduce results in one long training sequence.

Pitch

No response

Alternatives

No response

Additional context

No response

Checklist

• I have checked that there is no similar issue in the repo