How do I calculate perplexity?

[ShrimpLeopold]

Hi there! I need to check if my dialogue model's output is a valid English sentence. As far I'm concerned, it can be done by calculating perplexity of model's output.
I use transformer/generator model and create a agent with create_agent() function passing a model file into it.
I wonder whether I should create a Torch Generator Agent cause it has the forward method (the prolem is that I don't understand how to create it) or is there another way?

[stephenroller]

Easiest way is to follow our teacher tutorial to create a dataset containing the phrases you wish to evaluate, and use eval_model with that dataset.

[ShrimpLeopold]

Sorry, but the point is to measure PPL of output to new, unknown phrases