For Interview Perspective, What is more important knowing "how ML model training works/how deep learning data analysis works" or "knowing these things + each and every function we used/imported while training" ??
Each and every function that you import are made available by frameworks like (PyTorch, Tensorflow, SciPy…) to make building a solution quickly instead of worrying about re-writing commonly used code snippets.
Research oriented jobs would love to see, how well you understand the algorithms, your experimentation skills on how would improve on some approach that already exits, modifying the framework functions for you to experiment on your new approach.
Application/business oriented would like, how the analysis can improve or make their business model more efficient, getting predictions in real-time, being cost conscious(GPU renting ) , how will you overcome challenges (overfitting, prediction time…).
These are not clear cut bifurcation since the field is still evolving, know hows on both side of things would help.