How to train a DL model in Python when samples are 2d number arrays with different length?

I want to create a deep learning model, I know how to do so with simple samples where each row of dataset represents a sample, but my samples are 2D table of numbers. The features are the same in all samples but their length (rows) varies from sample to sample, like tables below.

Sample1:

Sample2:

Feature are the same, for example let's say feature 1 represents "Name" in all samples, feature 2 represents "city" and so on; but the number of rows in each sample varies.

I can not put away some rows to make the samples equal in length since they affect result. I also can not turn rows into new features to transform samples into 1D array of numbers because there are too many rows to do that.

1. I want to know do I need to somehow normalize the dimension of the samples? (if yes how can I do that?)
2. or Is there a deep learning model that can take inputs with different dimensions like my samples?

(I am using Python; I read that one way is to encode and decode the samples using LSTM seq-2-seq but I don't know if it is the right way to handle my dataset.)

Thanks in advance.