Self chat with two different models

[dianaglzrico]

Patch description
Allow self chat to be performed between two different models.

Testing steps
To self chat with two different models, add partner like so:

python parlai/scripts/self_chat.py -mf zoo:pretrained_transformers/model_poly/model -p zoo:blender/blender_90M/model

Self chat with same model, remains the same:

python parlai/scripts/self_chat.py -mf zoo:blender/blender_90M/model

Logs

$ py parlai/scripts/self_chat.py  -mf zoo:pretrained_transformers/model_poly/model -p zoo:blender/blender_90M/model
...
[context]: Hi!
14:09:55 WARN | [ Executing eval mode with a common set of fixed candidates (n = 131438). ]
[Polyencoder]: hey
   [TransformerGenerator]: hello , how are you today ? i just got back from a long day at work , so i ' m nervous .
[Polyencoder]: ahhh i remember those days , i am good though , stressing over work though
   [TransformerGenerator]: that ' s good to hear . do you have any other hobbies you like to do ?
[Polyencoder]: not really besides animals and exercising when i can ! you ?
   [TransformerGenerator]: i like to play video games and watch netflix . what kind of animals do you exercise ?
[Polyencoder]: all animals ! i love animals . do you have any hobbies ?
   [TransformerGenerator]: yes , i love playing video games . i also like to read . what ' s your favorite book ?
[Polyencoder]: that is a hard question . i would have to say the bible
   [TransformerGenerator]: i love the bible as well . it ' s a great book . what do you do for fun ?
-- end of episode --

$ py parlai/scripts/self_chat.py  -mf zoo:blender/blender_90M/model
...
[context]: Hi!
[TransformerGenerator_1]: hello , how are you today ? i just got back from a long day at work , so i ' m nervous .
   [TransformerGenerator_2]: i am nervous as well . what do you do for work ? i hope you have a good day !
[TransformerGenerator_1]: thank you ! i work at a bank . i hope your day goes as well as mine too !
   [TransformerGenerator_2]: thank you so much ! i hope it goes well for you too . i am sure you will have a great day !
[TransformerGenerator_1]: thank you , i ' ll try my best to make it go as smoothly as possible . do you have any hobbies ?
   [TransformerGenerator_2]: i like to play video games and watch netflix . i also like to read . what about you ?
[TransformerGenerator_1]: i love video games as well ! i ' ve been playing a lot of call of duty lately .
   [TransformerGenerator_2]: i haven ' t played that one , but i have heard good things about it . what is your favorite game ?
[TransformerGenerator_1]: my favorite game is black ops 3 . i have been playing it since it came out in 2013 .
   [TransformerGenerator_2]: that ' s a long time ago ! i have never played it . i will have to check it out .
-- end of episode --

[emilydinan]

nice job diana! one thing that I think is missing here is the ability to pass opts to the partner model if they differ from the primary model. one solution is to pass opt via a file. what do you think of this @stephenroller

[emilydinan]

great job diana! i have a few small requested changes below and then this will be ready to merge! 😄

[emilydinan]

just realized something and updated my comments

[github-actions]

Your PR contains a change to a task. Please paste the results of the following command into a comment:

python tests/datatests/test_new_tasks.py

[emilydinan]

just a couple really tiny changes and then we're there!

[emilydinan]

thank you for persisting through all of these iterations! looks great ⭐