add llm ai

src/classes/action.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 from abc import ABC, abstractmethod
 from typing import TYPE_CHECKING
 import random
+import json
 
 from src.classes.essence import Essence, EssenceType
 from src.classes.root import Root, corres_essence_type
@@ -48,6 +49,7 @@ class Move(DefineAction):
     """
     最基础的移动动作，在tile之间进行切换。
     """
+    COMMENT = "移动到某个相对位置"
     def execute(self, delta_x: int, delta_y: int) -> Event|NullEvent:
         """
         移动到某个tile
@@ -71,10 +73,13 @@ class MoveToRegion(DefineAction):
     """
     移动到某个region
     """
-    def execute(self, region: Region) -> Event|NullEvent:
+    COMMENT = "移动到某个区域"
+    def execute(self, region: Region|str) -> Event|NullEvent:
         """
         移动到某个region
         """
+        if isinstance(region, str):
+            region = self.world.map.region_names[region]
         cur_loc = (self.avatar.pos_x, self.avatar.pos_y)
         region_center_loc = region.center_loc
         delta_x = region_center_loc[0] - cur_loc[0]
@@ -89,6 +94,7 @@ class Cultivate(DefineAction):
     """
     修炼动作，可以增加修仙进度。
     """
+    COMMENT = "修炼，增进修为"
     def execute(self) -> Event|NullEvent:
         """
         修炼
@@ -116,6 +122,7 @@ class Breakthrough(DefineAction):
     """
     突破境界
     """
+    COMMENT = "尝试突破境界"
     def calc_success_rate(self) -> float:
         """
         计算突破境界的成功率
@@ -137,4 +144,12 @@ class Breakthrough(DefineAction):
         return Event(self.world.year, self.world.month, f"{self.avatar.name} 突破境界{res}")
 
 
-ALL_ACTION_CLASSES = [Move, Cultivate, Breakthrough, MoveToRegion]
+ALL_ACTION_CLASSES = [Move, Cultivate, Breakthrough, MoveToRegion]
+# 不包括Move
+ACTION_SPACE = [
+    # {"action": "Move", "params": {"delta_x": int, "delta_y": int}, "comment": Move.COMMENT},
+    {"action": "Cultivate", "params": {}, "comment": Cultivate.COMMENT},
+    {"action": "Breakthrough", "params": {}, "comment": Breakthrough.COMMENT},
+    {"action": "MoveToRegion", "params": {"region": "region_name"}, "comment": MoveToRegion.COMMENT},
+]
+ACTION_SPACE_STR = json.dumps(ACTION_SPACE, ensure_ascii=False) 

src/classes/ai.py

@@ -8,6 +8,8 @@ from abc import ABC, abstractmethod
 from src.classes.world import World
 from src.classes.tile import Region
 from src.classes.root import corres_essence_type
+from src.classes.action import ACTION_SPACE_STR
+from src.utils.llm import get_ai_prompt_and_call_llm
 
 class AI(ABC):
     """
@@ -23,12 +25,6 @@ class AI(ABC):
         """
         pass
 
-    # def create_event(self, world: World, content: str) -> Event:
-    #     """
-    #     创建事件
-    #     """
-    #     return Event(world.year, world.month, content)
-
 class RuleAI(AI):
     """
     规则AI
@@ -69,4 +65,14 @@ class LLMAI(AI):
         """
         决定做什么
         """
-        pass
+        action_space_str = ACTION_SPACE_STR
+        avatar_infos_str = str(self.avatar)
+        regions_str = "\n".join([str(region) for region in world.map.regions.values()])
+        dict_info = {
+            "action_space": action_space_str,
+            "avatar_infos": avatar_infos_str,
+            "regions": regions_str
+        }
+        res = get_ai_prompt_and_call_llm(dict_info)
+        action_name, action_params = res["action_name"], res["action_params"]
+        return action_name, action_params

src/classes/avatar.py

@@ -12,7 +12,7 @@ from src.classes.cultivation import CultivationProgress, Realm
 from src.classes.root import Root
 from src.classes.age import Age
 from src.utils.strings import to_snake_case
-from src.classes.ai import AI, RuleAI
+from src.classes.ai import AI, RuleAI, LLMAI
 
 class Gender(Enum):
     MALE = "male"
@@ -52,9 +52,17 @@ class Avatar:
         在Avatar创建后自动绑定基础动作和AI
         """
         self.tile = self.world.map.get_tile(self.pos_x, self.pos_y)
-        self.ai = RuleAI(self)
+        self.ai = LLMAI(self)
+        # self.ai = RuleAI(self)
         self._bind_basic_actions()
 
+    def __str__(self) -> str:
+        """
+        获取avatar的详细信息
+        尽量多打一些，因为会用来给LLM进行决策
+        """
+        return f"Avatar(id={self.id}, 性别={self.gender}, 年龄={self.age}, name={self.name}, 区域={self.tile.region.name}, 灵根={self.root.value}, 境界={self.cultivation_progress})"
+
     def _bind_basic_actions(self):
         """
         绑定基础动作，如移动等

src/classes/cultivation.py

@@ -148,4 +148,7 @@ class CultivationProgress:
         """
         检查是否可以突破
         """
-        return self.level in level_to_break_through.keys()
+        return self.level in level_to_break_through.keys()
+
+    def __str__(self) -> str:
+        return f"{self.realm.value}{self.stage.value}({self.level}级)。可以突破：{self.can_break_through()}"

src/classes/tile.py

@@ -45,6 +45,15 @@ class Region():
     def __post_init__(self):
         self.id = next(region_id_counter)
 
+    def __str__(self) -> str:
+        return f"区域。名字：{self.name}，描述：{self.description}，最浓的灵气：{self.get_most_dense_essence()}， 灵气值：{self.get_most_dense_essence_value()}"
+
+    def get_most_dense_essence(self) -> EssenceType:
+        return max(self.essence.density.items(), key=lambda x: x[1])[0]
+    
+    def get_most_dense_essence_value(self) -> int:
+        most_dense_essence = self.get_most_dense_essence()
+        return self.essence.density[most_dense_essence]
 
     def __hash__(self) -> int:
         return hash(self.id)
@@ -74,7 +83,8 @@ class Map():
     """
     def __init__(self, width: int, height: int):
         self.tiles = {}
-        self.regions = {}
+        self.regions = {}      # region_id -> region
+        self.region_names = {} # region_name -> region
         self.width = width
         self.height = height
 
@@ -101,6 +111,7 @@ class Map():
         region.center_loc = center_loc
         region.area = len(locs)
         self.regions[region.id] = region
+        self.region_names[name] = region
         return region
 
     def get_center_locs(self, locs: list[tuple[int, int]]) -> tuple[int, int]:

src/run/run.py

@@ -93,7 +93,7 @@ def main():
     sim = Simulator(world)
     
     # 创建角色，传入当前年份确保年龄与生日匹配
-    sim.avatars.update(make_avatars(world, count=14, current_year=world.year))
+    sim.avatars.update(make_avatars(world, count=2, current_year=world.year))
 
     front = Front(
         simulator=sim,

src/utils/config.py

@@ -32,6 +32,10 @@ def load_config():
     
     # 合并配置，local_config优先级更高
     config = OmegaConf.merge(base_config, local_config)
+
+    # 把paths下的所有值pathlib化
+    for key, value in config.paths.items():
+        config.paths[key] = Path(value)
     
     return config
 

src/utils/io.py

@@ -0,0 +1,8 @@
+from pathlib import Path
+
+def read_txt(path: Path) -> str:
+    """
+    读入中文txt文件
+    """
+    with open(path, "r", encoding="utf-8") as f:
+        return f.read()

src/utils/llm.py

@@ -1,7 +1,10 @@
 from litellm import completion
 from langchain.prompts import PromptTemplate
+from pathlib import Path
+import json
 
 from src.utils.config import CONFIG
+from src.utils.io import read_txt
 
 def get_prompt(template: str, infos: dict) -> str:
     """
@@ -31,4 +34,23 @@ def call_llm(prompt: str) -> str:
     )
     
     # 返回生成的内容
-    return response.choices[0].message.content
+    return response.choices[0].message.content
+
+def get_prompt_and_call_llm(template_path: Path, infos: dict) -> str:
+    """
+    根据模板，获取提示词，并调用LLM
+    """
+    template = read_txt(template_path)
+    prompt = get_prompt(template, infos)
+    res = call_llm(prompt)
+    json_res = json.loads(res)
+    print(f"prompt = {prompt}")
+    print(f"res = {res}")
+    return json_res
+
+def get_ai_prompt_and_call_llm(infos: dict) -> dict:
+    """
+    根据模板，获取提示词，并调用LLM
+    """
+    template_path = CONFIG.paths.templates / "ai.txt"
+    return get_prompt_and_call_llm(template_path, infos)

static/config.yml

@@ -1,4 +1,7 @@
 llm:
   # 填入litellm支持的model name和key
   model_name: 
-  key:
+  key:
+
+paths:
+  templates: static/templates/

static/templates/ai.txt

@@ -0,0 +1,18 @@
+
+
+
+你是一个决策者，这是一个修仙的仙侠世界，你负责来决定一些NPC的下一步行为。
+每个角色均拥有的动作空间和需要的参数为：
+{action_space}
+世界地图上存在的区域为：
+{regions}
+你需要进行决策的NPC的基本信息为：
+{avatar_infos}
+
+注意，只返回json格式的动作
+返回格式：
+{{
+    "avatar_id": ...,
+    "action_name": ...,
+    "action_params": ...,
+}}