Adding dummy option for testing

[GivAlz]

Hi everyone! First I'd like to thank OpenAI for their wonderful products and push forward in the research!

I'm writing because there is one theme that I find painful when using OpenAI.

That's testing.

Sometimes it makes no sense to call the OpenAI model during tests, but this means I have to build a dummy class by hand, which is sort of time consuming.

Before building it as a separate, standalone pipy package, I was wondering if I could try and add it directly to the openai python package.

My idea would be the following:

• adding a flag, something like client = openai.OpenAI(dummy=True)
• adding a method (for the seq2seq models) like client.load_dummy_answers(default: str, Optional[Dict[str, str]]), where one sets up a default answer and a dictionary where key is prompt and the value is the dummy response
• the other parameters returned by the models would default dummies (except maybe for the time and date of generations)

(of course I would add this, for now, just to text2text models...in theory this could be extended to other models too).

My idea is to do everything in package...but I'm sure you could do something similar adding a "dummy" model to the list (but then people would still need to use the internet connection for testing).