Getting the output of the final layer

[mitramir55]

Hi @jalammar,

Currently, In a research project, I'm trying to give one token to GPT2 and then output the exact ranking you get in the layer_predictions function. Eventually, I intend to combine this output with a translation model output (both of the outputs should have the length of the vocabulary) and then get one token chosen by linearly combining both models ranking (this method is called shallow fusion in NMT).

Is it possible that I use your package to do this? I'd really appreciate your help and advice on how this can be possible!