Tutorial for training loop w/o assistant

[mathisrichter]

User story

As a user, I want to have a tutorial that explains how to implement a training loop without the assistant to clarify how Pytorch training is done.

Note: Intel NCL will not actively work on this story. If someone in the community would like to take this on, feel free to contact us or open a PR. One idea would be to extract the part of the MNIST tutorial that deals with training into a dedicated tutorial. See comments below.

[mathisrichter]

@bamsumit Could you add to the story what the benefit would be to be able to not use the assistant? Does this unlock more features or make users more flexible in implementing the training loop?

Also, could you add a priority label? It is probably not urgent but if this is core feature, I would say clear documentation is important. If the feature is more exotic, I would think it's not so important.

[bamsumit]

@mathisrichter The MNIST training tutorial already shows how the training can be done without the assistant. It is the same steps as training a standard PyTorch network, so it would not add much. Not using the assistant would perhaps clarify how PyTorch training is done, nothing specific to lava-dl. I think it will do a good job of educating people and make it easier to tweak some aspects of the training. I clearly remember many people found having the loops in the example jarring (I personally don't think so) when we reviewed the tutorials, which actually lead to the work on assistants.

I would keep this as a very low priority for now.

[mathisrichter]

Great, thanks @bamsumit for looking into this. We can keep this feature request around as a potential community project and not work on it ourselves. I labeled the story accordingly.