Training data == source code, training algorithm == compiler, model weights == compiled binary.

Training algorithm is the programmer, weights are the code that you run in an interpreter