Spaces:
Running
Running
File size: 1,933 Bytes
8efd70f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 | """
client.py — Client-side EnvClient for SENTINEL.
Used by training scripts (TRL GRPOTrainer) to talk to a running SENTINEL
server. Wraps HTTP calls in typed Pydantic models.
Usage (training notebook):
from sentinel.client import SentinelEnv, ResponderAction, OverseerAction
env = SentinelEnv(base_url="https://Elliot89-sentinel.hf.space")
obs = env.reset(task_id="war_room", seed=42).observation
step = env.step(OverseerAction(decision="block", justification="rm -rf on prod db"))
"""
from __future__ import annotations
from typing import Any
try:
from openenv.core.env_client import EnvClient # type: ignore
except Exception: # pragma: no cover — lets the repo import in dev without openenv-core
EnvClient = object # type: ignore
from models import (
Action,
DualReward,
EpisodeState,
Observation,
OverseerAction,
ResponderAction,
StepResult,
)
class SentinelEnv(EnvClient): # type: ignore[misc]
"""Typed client for the SENTINEL OpenEnv."""
Action = Action
Observation = Observation
State = EpisodeState
def _step_payload(self, action: Action | ResponderAction | OverseerAction) -> dict:
if isinstance(action, ResponderAction):
return {"role": "responder", "responder": action.model_dump()}
if isinstance(action, OverseerAction):
return {"role": "overseer", "overseer": action.model_dump()}
return action.model_dump()
def _parse_result(self, payload: dict[str, Any]) -> StepResult:
return StepResult(
observation=Observation(**payload["observation"]),
reward=DualReward(**payload["reward"]),
done=payload.get("done", False),
info=payload.get("info", {}),
)
def _parse_state(self, payload: dict[str, Any]) -> EpisodeState:
return EpisodeState(**payload)
|