https://github.com/minimaxir/gpt-2-simple/blob/ca6bc61d958fd4c474af9a412ace27279b88dd90/gpt_2_simple/src/encoder.py#L8

According to the docs the lru_cache decorator uses memozation to speed up a function call when it's called with the same arguments.

But this function doesn't need any arguments...

It would be great to have instructions on how to train a language model from scratch - not just loading the paper's model.

Hi,
When we try to tokenize the following sentence:

If we use spacy

a = spacy.load('en_core_web_lg')

doc = a("I like the link http://www.idph.iowa.gov/ohds/oral-health-center/coordinator")

list(doc)

We got

[I, like, the, link, http://www.idph.iowa.gov, /, ohds, /, oral, -, health, -, center, /, coordinator]

But if we use the Spacy transformer tokenizer:

At the moment if the digital twin model does not have the plugins built, we can not communicate them them thus we get the following error,
Timeout communicating with flight control plugin.
It would be helpful if the error message included a message reminding the user to check to make sure the plugin is built, or better yet do the check automatically.

Similarly as we do for other projects (example) we should add a CONTRIBUTING file. It would be very helpful for external contributors to understand how to propose new changes.

Currently the ConsolePrinter spams states, references and reward information step by step, but this is no useful visualization.

A more readable and descriptive concept must be elaborated.

Moreover, documentation is missing for the ConsolePrinter.