add bridge

[Muhtasham]

No description provided.

[john-b-yang]

@Muhtasham this is fantastic - thank u so much.

An immediate feedback point: For CodeClash, the arena engine + code (what you put under bridge/examples and bridge/game_server is typically placed in a different repository (for instance, RobotRumble)

Can you do the same for Bridge? We want to keep the arena code separate from the repository code for better maintainability + enable customization more easily on a per-arena basis.

Let me know if you have any questions here. I'd reference an existing .Dockerfile to also see what updates are needed. Some random notes:

• Put any arena/game-specific code into a separate repository
• Your .Dockerfile should clone the github repository
• Your codeclash/arenas/bridge/bridge.py looks great, I didn't take a close look, but if there's a need, make sure it's grounded in the paths of the new, separate repository.

Once you do these fixes, let me know and I'll give the arena a couple runs and see how it goes.

Great initial work! 🙏🏼

[john-b-yang]

Ok implementation is fantastic, thank you so much @Muhtasham!

I made one update to the original repository, I moved examples/random_agent.py to just be the default bridge_agent.py. For all of the arenas, one of the things we ensured in the original paper is that starter codebase contains an already-functioning implementation of the bot. This is a bit for "fairness" purposes (we give the LM a working implementation, if it breaks it, that's the model's fault), and also empirical to enable investigation of relatively "interesting" behaviors ("uninteresting" = the model just fumbles around trying to figure out how to put together an initial bot for multiple rounds, although this could be interesting to investigate).

I think this PR is ready to merge! Just approved - will let you take a last pass, and then whenever you feel comfortable, feel free to merge!