Spaces:
Sleeping
Sleeping
File size: 2,544 Bytes
fb78c46 e27dbae fb78c46 | 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 59 60 61 62 63 64 65 66 | from typing import Dict, Any
from .models import Observation, Action, Reward
from .utils import generate_ticket
from .tasks import get_task
from .reward import calculate_reward
from .graders import grade_easy, grade_medium, grade_hard
class SupportDeskEnv:
def __init__(self):
self.state_data = None
self.expected_category = None
self.task_config = None
async def reset(self, level: str = "medium") -> Observation:
self.task_config = get_task(level)
ticket = generate_ticket(self.task_config.level)
self.expected_category = ticket["category"]
self.state_data = Observation(
ticket_id=ticket["id"],
user_message=ticket["message"],
sentiment=ticket["sentiment"],
history=[{"role": "user", "content": ticket["message"]}],
step_count=0,
task_level=self.task_config.level
)
return self.state_data
async def step(self, action: Action) -> Dict[str, Any]:
if not self.state_data:
await self.reset()
self.state_data.step_count += 1
self.state_data.history.append({"role": "agent", "content": action.response or ""})
# Pass max_steps to calculate_reward for the efficiency bonus
reward = calculate_reward(self.state_data, action, self.expected_category, self.task_config.max_steps)
done = action.resolve or action.escalate or self.state_data.step_count >= self.task_config.max_steps
# Grading based on level
if self.task_config.level == "easy":
task_score = grade_easy(action, self.expected_category)
elif self.task_config.level == "medium":
task_score = grade_medium(action, self.expected_category)
else:
task_score = grade_hard(action, self.state_data, self.expected_category)
reward.metrics["grader_score"] = task_score
if not done:
self.state_data.user_message = "Can you explain more?"
self.state_data.history.append({"role": "user", "content": self.state_data.user_message})
return {
"observation": self.state_data.model_dump(),
"reward": reward.model_dump(),
"done": done,
"info": {"expected_category": self.expected_category, "task_score": task_score}
}
async def state(self) -> Observation:
if not self.state_data:
await self.reset()
return self.state_data
|