Clarify what is `ObsType`: is it the type for a single observation, or a list of observations?

[etnbrd]

When reading the Env implementation and the documentation, it's unclear if ObsType should be a single observation or a list of observations.

The EnvStepOutput type have a field observations (plural) typed ObsType:

SkyRL/skyrl-gym/skyrl_gym/core.py

Line 13 in b1107ed

observations: ObsType

At the same time, the step method documents that it returns observation (singular):

SkyRL/skyrl-gym/skyrl_gym/core.py

Line 45 in b1107ed

observation (ObsType): The resulting observation after executing the action.

In the BaseTextEnv implementation Env is initialised with ObsType being a str:

SkyRL/skyrl-gym/skyrl_gym/envs/base_text_env.py

Line 17 in b1107ed

class BaseTextEnv(Env[str, str]):

But the step methods return List[Dict[str, str]] in the observations field:

SkyRL/skyrl-gym/skyrl_gym/envs/base_text_env.py

Line 10 in b1107ed

observations: List[Dict[str, str]] # OpenAI API Messages Format

I'd be happy to provide a PR to unify these types and documentation, but I'm not sure what ObsType should contain.

[tyler-griggs]

Hey @etnbrd great catch, thanks for spotting this.

I believe inbase_text_env.py we should update both the observations field and the ObsType to be ConversationType (which is defined at the top of the file). This way, the observations are simply the standard OpenAI SDK Conversation format. Then, I think the step method documentation in skyrl_gym/core.py should be updated to change observation to the plural observations. What do you think?

[etnbrd]

Let's go: #207

[tyler-griggs]

Great, thank you!