Seeds to make results reproducible

There is some randomness in the code, leading to different results for every execution. For example, when we train the neural network in task 7, the loss for each epoch vary. Should we use seeds so the results are reproducible?

<img width="439" alt="image" src="https://github.com/MultiX-Amsterdam/text-data-module/assets/149786124/7a55f08e-4c92-4ad9-ad42-f3f2635e10d1">
