64 lines
2.2 KiB
Python
64 lines
2.2 KiB
Python
from typing import Any, Dict, List, Optional
|
|
from src.utils.llm import call_llm_with_template
|
|
from src.classes.avatar import Avatar
|
|
import json
|
|
|
|
async def make_decision(
|
|
avatar: Avatar,
|
|
context_desc: str,
|
|
options: List[Dict[str, Any]]
|
|
) -> str:
|
|
"""
|
|
让角色在多个选项中做出单选决策。
|
|
|
|
Args:
|
|
avatar: 做出决策的角色
|
|
context_desc: 决策背景描述
|
|
options: 选项列表，每个选项是一个字典，必须包含 'key' 和 'desc' 字段
|
|
例如: [{'key': 'A', 'desc': '...'}, {'key': 'B', 'desc': '...'}]
|
|
|
|
Returns:
|
|
str: AI 选择的选项 Key (如 'A' 或 'B')
|
|
"""
|
|
# 1. 获取角色信息 (详细模式)
|
|
avatar_infos = str(avatar.get_info(detailed=True))
|
|
|
|
# 2. 格式化选项字符串
|
|
choices_list = [f"{opt.get('key', '')}: {opt.get('desc', '')}" for opt in options]
|
|
choices_str = "\n".join(choices_list)
|
|
full_choices_str = f"【当前情境】：{context_desc}\n\n{choices_str}"
|
|
|
|
# 3. 调用 AI
|
|
result = await call_llm_with_template(
|
|
"single_choice",
|
|
avatar_infos=avatar_infos,
|
|
choices=full_choices_str
|
|
)
|
|
|
|
# 4. 解析结果
|
|
choice = ""
|
|
if isinstance(result, dict):
|
|
choice = result.get("choice", "").strip()
|
|
elif isinstance(result, str):
|
|
clean_result = result.strip()
|
|
# 尝试解析可能包含 markdown 的 json
|
|
if "{" in clean_result and "}" in clean_result:
|
|
try:
|
|
# 提取可能的 json 部分
|
|
start = clean_result.find("{")
|
|
end = clean_result.rfind("}") + 1
|
|
json_str = clean_result[start:end]
|
|
data = json.loads(json_str)
|
|
choice = data.get("choice", "").strip()
|
|
except (json.JSONDecodeError, ValueError):
|
|
# 如果 JSON 解析失败，直接看字符串内容是否就是选项 key
|
|
choice = clean_result
|
|
else:
|
|
choice = clean_result
|
|
|
|
# 验证 choice 是否在 options key 中
|
|
valid_keys = {opt["key"] for opt in options}
|
|
assert choice in valid_keys, f"choice {choice} not in valid_keys {valid_keys}"
|
|
return choice
|
|
|