Well, I know this is really hard at beginning. I share several opinions hoping to help you.

1. Whatever you read before which you need to have a very solid command of the knowledge (models) on the text book. This is the very fundamental pre-knowledge that will help you to understand some further advanced ones.

2. After you have a good knowing of one model, you can go searching the original paper of it. The model on the text book are usually very well known and is revised or articulated by many authors. Nevertheless, the original paper sometimes is more complicated and in many times you may find few similarities with that appeared on the text book. It usually include the ideas and backgrounds of how they come up with the model, and detailed derivation. But since you have known the model, you can handle with it. Sometimes it's tough, but you can do it.

3. After you are very familiar with the traditional models, you can get to some new models in the academic frontier. These papers are much harder perhaps in the algorithm or something else, but the main idea behind them unexceptionally deviates from those traditional models. So just catch the main point.

Some tips when reading some complicated paper.

1) When you try to catch the main point of this paper, try to do it in this way. First, the motivation is, from my point of view, the most important, nothing is more important than the reason why they write this paper. A good paper should have a good motivation. Many bad papers are not well orgnized and just showing how they have a good knowledge of math :-).

2) Whenever encountering very complex mathematical derivation, just try to understand to know what it does. If you don't know the math he uses, just assume you know it and then goes down, don't think that if you don't know the math you can not know the model. In most cases, math is just a tool to reach destination, we care more about the destinations.

The traditional model I mentioned includes something like Black-Scholes, Vasicek, CIR, HJM, LMM, Hull-White SV model, Heston SV (SV=Stochastic Volatility).

For example, Heston's SV, he uses Fourier transform to solve the option pricing problem. So you must be very scared to read it if you have learn nothing about Fourier transform. But try to read it. You see first he will discuss, make a conclusion of the SV model that have already existed, including make some comments, so that in the introduction part, you actually can learn the whole history of many related researches and many brief description of different models. Then, you see how does he set his model and how does he solve it. Guess the solution to be very similar to Black-Scholes (option's price form should be very close to BS, this idea should be caught, because it is a general pattern). Then the detailed part of solving it. If you can't understand Fourier transform, just ignore it. You know that somehow this problem can be solved.

As to how detailed should you involve in the paper, it depends. For the papers I mentioned above, it is beneficial to catch every single word. But papers also vary very much in quality, many papers are too bad that you can just drag them in the trash can. Some papers have good ideas, but the body is terrible. The good paper is usually referred by many authors, you can search them in this way. For these papers, you do not need to follow all the details, just do as what I have mentioned. But if you decide to implement and method in these papers, you then should catch many detials.

It is a complicated and  hard work, and I only mention a very small part of it. Hope it can help you.

best,

Thanks a lot. Though I am not good at English reading, I will try my best.