|
|
from abc import ABC, abstractmethod |
|
|
from typing import Literal, Dict, Tuple |
|
|
|
|
|
class BaseEnv(ABC): |
|
|
ENV_CARD: Literal["STATIC", "DYNAMIC"] = None |
|
|
|
|
|
def __init__(self, config): |
|
|
self.config = config |
|
|
|
|
|
@classmethod |
|
|
@abstractmethod |
|
|
def _accuracy_reward(cls, **kwargs): |
|
|
... |
|
|
|
|
|
@classmethod |
|
|
@abstractmethod |
|
|
def _format_reward(cls, **kwargs): |
|
|
... |
|
|
|
|
|
@classmethod |
|
|
def get_reward_func(cls, func_name: Literal["accuracy", "format"]): |
|
|
if func_name == "accuracy": |
|
|
return cls._accuracy_reward |
|
|
|
|
|
elif func_name == "format": |
|
|
return cls._format_reward |
|
|
|
|
|
else: |
|
|
raise ValueError(f"Unsupported reward func: {func_name}") |
|
|
|
|
|
class StaticEnv(BaseEnv): |
|
|
ENV_CARD = "STATIC" |
|
|
|
|
|
|
|
|
class DynamicEnv(BaseEnv): |
|
|
ENV_CARD = "DYNAMIC" |
|
|
|
|
|
@abstractmethod |
|
|
def set_env(self, task_config: Dict) -> Tuple[str, str]: |
|
|
... |
|
|
|
|
|
@classmethod |
|
|
@abstractmethod |
|
|
def preprocess_action(self, action: str) -> str: |
|
|
... |
|
|
|
|
|
@abstractmethod |
|
|
def step(self, action: str) -> Tuple[str, bool]: |
|
|
... |
|
|
|
|
|
@abstractmethod |
|
|
def feedback(self) -> Tuple[float, bool]: |
|
|
... |
