Create a unified API between PPO and ILQL

[LouisCastricato]

Currently ILQL was implemented more or less in a vacuum of our PPO implementation. As such, ILQL has features that our PPO implementation needs. This includes

• Removing the dependency on GPT2 for PPO. This is currently being done here, but we can create a more standardized approach
• Adding the ability to pass custom reward functions to the orchestrator. This is an absolute must have.
• Standardized naming conventions between functions.

[LouisCastricato]

@reciprocated @Dahoas no rush on this but whenever you guys get a chance we should discuss this in this issue.

[LouisCastricato]

Fixed the first point with #11

All that remains is standardizing naming conventions.

[maxreciprocate]

On the topic of unified api, to encourage new users I vote for simplifying down to

import trlx
# for ppo
trlx.train(model_path, reward_fn)
# for ilql
trlx.train(model_path, samples, ratings)

[LouisCastricato]

I like that, good idea