Training on a tensors without common DataLoader

I am sorry if my question sounds too noobish. I am beginner both in python and in torch.
So I managed to iterate through my dataset by a custom way and created a function which is resulting
me an a = InputDataTensor, OutputDataTensor
My question is how to train a model based on usage of my function which is on every call gives me new input data of tensor[10,1] and corresponding output data of [100,1] (sizes I will adjust in future).
The thing is that my dataset is just an input and corresponding output. I am not working with any prebuilt datasets from torch and guides on dataloaders are just confusing me with always working over MNIST.

Have you had the chance look at the Quickstart guide? This section shows how you can pass in data to train your model.

This guide is showing work with dataloaders and MNIST, which I wanted to avoid. Thanks anyway. I am thinking now towards assembling conventional dataset from my func. so I just can swith to dataloaders.

If you do not wish to use the DataLoader, you can replace enumerate(dataloader) from the tutorial with your own iterator object.

Let me know if there is any issue with that.

Thanks. I have a solution now. I just automated creation of a guide-like dataset from my data and now I just follow guides with a regular dataloader which is able now to load data I need. It sounds so simple, but was a hard to come up with for such a newibe like me.