67 lines
2.1 KiB
Python
67 lines
2.1 KiB
Python
"""
|
||
ExtractionAgent：从用户消息中提取 5-stage 状态与 slots。
|
||
对应现有逻辑：get_state_extraction_prompt + JSON 解析
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import json
|
||
from dataclasses import dataclass
|
||
from typing import Any, Dict
|
||
|
||
from app.core.logging import get_logger
|
||
from app.features.memoir.memoir_images.json_payload import extract_json_payload
|
||
|
||
from app.agents.memoir.prompts import get_state_extraction_prompt
|
||
|
||
logger = get_logger(__name__)
|
||
|
||
|
||
@dataclass
|
||
class ExtractionResult:
|
||
"""状态提取结果"""
|
||
detected_stage: str
|
||
slots: Dict[str, str]
|
||
|
||
|
||
class ExtractionAgent:
|
||
"""从用户消息中提取 detected_stage 和 slots"""
|
||
|
||
def extract(
|
||
self,
|
||
user_message: str,
|
||
current_stage: str,
|
||
stage_slots: Dict[str, Any],
|
||
llm: Any,
|
||
) -> ExtractionResult:
|
||
"""
|
||
提取结构化信息并判断阶段。
|
||
llm 需支持 .invoke(prompt) 同步调用（Celery 任务内使用）。
|
||
"""
|
||
detected_stage = current_stage
|
||
extracted_slots: Dict[str, str] = {}
|
||
|
||
if not llm:
|
||
return ExtractionResult(detected_stage=detected_stage, slots=extracted_slots)
|
||
|
||
try:
|
||
prompt = get_state_extraction_prompt(
|
||
user_message=user_message,
|
||
current_stage=current_stage,
|
||
stage_slots={
|
||
k: v.model_dump() if hasattr(v, "model_dump") else v
|
||
for k, v in (stage_slots or {}).items()
|
||
},
|
||
)
|
||
response = llm.invoke(prompt)
|
||
parsed = json.loads(extract_json_payload(response.content))
|
||
detected_stage = parsed.get("detected_stage", detected_stage)
|
||
raw_slots = parsed.get("slots", {}) or {}
|
||
extracted_slots = {
|
||
k: v if isinstance(v, str) else str(v)
|
||
for k, v in raw_slots.items()
|
||
}
|
||
except (json.JSONDecodeError, Exception) as e:
|
||
logger.warning("ExtractionAgent LLM 解析失败: %s", e)
|
||
|
||
return ExtractionResult(detected_stage=detected_stage, slots=extracted_slots)
|