refactor llm config
This commit is contained in:
@@ -210,8 +210,57 @@ def serialize_phenomenon(phenomenon) -> Optional[dict]:
|
||||
"effect_desc": effect_desc
|
||||
}
|
||||
|
||||
def check_llm_connectivity() -> tuple[bool, str]:
|
||||
"""
|
||||
检查 LLM 连通性
|
||||
|
||||
Returns:
|
||||
(是否成功, 错误信息)
|
||||
"""
|
||||
try:
|
||||
from src.utils.llm.config import LLMMode, LLMConfig
|
||||
|
||||
normal_config = LLMConfig.from_mode(LLMMode.NORMAL)
|
||||
fast_config = LLMConfig.from_mode(LLMMode.FAST)
|
||||
|
||||
# 检查配置是否完整
|
||||
if not normal_config.api_key or not normal_config.base_url:
|
||||
return False, "LLM 配置不完整:请填写 API Key 和 Base URL"
|
||||
|
||||
if not normal_config.model_name:
|
||||
return False, "LLM 配置不完整:请填写智能模型名称"
|
||||
|
||||
# 判断是否需要测试两次
|
||||
same_model = (normal_config.model_name == fast_config.model_name and
|
||||
normal_config.base_url == fast_config.base_url and
|
||||
normal_config.api_key == fast_config.api_key)
|
||||
|
||||
if same_model:
|
||||
# 只测试一次
|
||||
print(f"检测 LLM 连通性(单模型): {normal_config.model_name}")
|
||||
success, error = test_connectivity(LLMMode.NORMAL, normal_config)
|
||||
if not success:
|
||||
return False, f"连接失败:{error}"
|
||||
else:
|
||||
# 测试两次
|
||||
print(f"检测智能模型连通性: {normal_config.model_name}")
|
||||
success, error = test_connectivity(LLMMode.NORMAL, normal_config)
|
||||
if not success:
|
||||
return False, f"智能模型连接失败:{error}"
|
||||
|
||||
print(f"检测快速模型连通性: {fast_config.model_name}")
|
||||
success, error = test_connectivity(LLMMode.FAST, fast_config)
|
||||
if not success:
|
||||
return False, f"快速模型连接失败:{error}"
|
||||
|
||||
return True, ""
|
||||
|
||||
except Exception as e:
|
||||
return False, f"连通性检测异常:{str(e)}"
|
||||
|
||||
def init_game():
|
||||
"""初始化游戏世界,逻辑复用自 src/run/run.py"""
|
||||
|
||||
print("正在初始化游戏世界...")
|
||||
game_map = load_cultivation_world_map()
|
||||
world = World(map=game_map, month_stamp=create_month_stamp(Year(100), Month.JANUARY))
|
||||
@@ -268,6 +317,24 @@ def init_game():
|
||||
game_instance["sim"] = sim
|
||||
print("游戏世界初始化完成!")
|
||||
|
||||
# ===== LLM 连通性检测(在 simulator 运行前)=====
|
||||
print("正在检测 LLM 连通性...")
|
||||
success, error_msg = check_llm_connectivity()
|
||||
|
||||
if not success:
|
||||
print(f"[警告] LLM 连通性检测失败: {error_msg}")
|
||||
print("[警告] Simulator 已暂停,等待配置 LLM...")
|
||||
game_instance["llm_check_failed"] = True
|
||||
game_instance["llm_error_message"] = error_msg
|
||||
game_instance["is_paused"] = True
|
||||
print("等待前端连接并配置 LLM...")
|
||||
else:
|
||||
print("LLM 连通性检测通过 ✓")
|
||||
game_instance["llm_check_failed"] = False
|
||||
game_instance["llm_error_message"] = ""
|
||||
game_instance["is_paused"] = False
|
||||
# ===== LLM 检测结束 =====
|
||||
|
||||
async def game_loop():
|
||||
"""后台自动运行游戏循环"""
|
||||
print("后台游戏循环已启动...")
|
||||
@@ -455,6 +522,17 @@ print(f"Web dist path: {WEB_DIST_PATH}")
|
||||
@app.websocket("/ws")
|
||||
async def websocket_endpoint(websocket: WebSocket):
|
||||
await manager.connect(websocket)
|
||||
|
||||
# ===== 检查 LLM 状态并通知前端 =====
|
||||
if game_instance.get("llm_check_failed", False):
|
||||
error_msg = game_instance.get("llm_error_message", "LLM 连接失败")
|
||||
await websocket.send_json({
|
||||
"type": "llm_config_required",
|
||||
"error": error_msg
|
||||
})
|
||||
print(f"已向客户端发送 LLM 配置要求: {error_msg}")
|
||||
# ===== 检测结束 =====
|
||||
|
||||
try:
|
||||
while True:
|
||||
# 保持连接活跃,接收客户端指令(目前暂不处理复杂指令)
|
||||
@@ -1024,17 +1102,22 @@ def test_llm_connection(req: TestConnectionRequest):
|
||||
model_name=req.model_name
|
||||
)
|
||||
|
||||
success = test_connectivity(config=config)
|
||||
success, error_msg = test_connectivity(config=config)
|
||||
|
||||
if success:
|
||||
return {"status": "ok", "message": "连接成功"}
|
||||
else:
|
||||
raise HTTPException(status_code=400, detail="连接失败")
|
||||
# 返回 400 错误并附带详细的错误信息
|
||||
raise HTTPException(status_code=400, detail=error_msg)
|
||||
except HTTPException:
|
||||
# 重新抛出 HTTPException
|
||||
raise
|
||||
except Exception as e:
|
||||
# 其他未预期的错误
|
||||
raise HTTPException(status_code=500, detail=f"测试出错: {str(e)}")
|
||||
|
||||
@app.post("/api/config/llm/save")
|
||||
def save_llm_config(req: LLMConfigDTO):
|
||||
async def save_llm_config(req: LLMConfigDTO):
|
||||
"""保存 LLM 配置"""
|
||||
try:
|
||||
# 1. Update In-Memory Config (Partial update)
|
||||
@@ -1075,6 +1158,24 @@ def save_llm_config(req: LLMConfigDTO):
|
||||
|
||||
OmegaConf.save(conf, local_config_path)
|
||||
|
||||
# ===== 如果之前 LLM 连接失败,现在恢复运行 =====
|
||||
if game_instance.get("llm_check_failed", False):
|
||||
print("检测到之前 LLM 连接失败,正在恢复 Simulator 运行...")
|
||||
|
||||
# 清除失败标志并恢复运行
|
||||
game_instance["llm_check_failed"] = False
|
||||
game_instance["llm_error_message"] = ""
|
||||
game_instance["is_paused"] = False
|
||||
|
||||
print("Simulator 已恢复运行 ✓")
|
||||
|
||||
# 通知所有客户端刷新
|
||||
await manager.broadcast({
|
||||
"type": "game_reinitialized",
|
||||
"message": "LLM 配置成功,游戏已恢复运行"
|
||||
})
|
||||
# ===== 恢复运行结束 =====
|
||||
|
||||
return {"status": "ok", "message": "配置已保存"}
|
||||
except Exception as e:
|
||||
import traceback
|
||||
@@ -1113,22 +1214,6 @@ def api_save_game(req: SaveGameRequest):
|
||||
if not world or not sim:
|
||||
raise HTTPException(status_code=503, detail="Game not initialized")
|
||||
|
||||
# 这里的 existed_sects 需要从 world 或者 sim 中获取,目前简单起见,
|
||||
# 我们可以遍历地图上的宗门总部,或者如果全局有保存最好。
|
||||
# 由于 init_game 只有一次,我们需要从 world 中反推 active sects
|
||||
# 但 save_game 签名里的 existed_sects 主要是为了记录 id。
|
||||
# 实际上 world.map.regions 中包含了宗门总部信息。
|
||||
# 或者更简单的:直接从 sects_by_id 取所有? 不太对。
|
||||
# 让我们看看 save_game 实现:它主要是存 id。
|
||||
# 我们可以传入空列表,如果在 load 时能容忍的话。
|
||||
# 实际上 load_game 里:existed_sects = [sects_by_id[sid] for sid in existed_sect_ids]
|
||||
# 所以 save 时如果不传,load 时就拿不到。
|
||||
# 临时方案:遍历所有宗门,如果它有领地或者有人,就算存在。
|
||||
# 或者更粗暴:CONFIG.game.sect_num 如果没变,可以不管。
|
||||
# 最好是 world 对象上能挂载 existed_sects。
|
||||
# 暂时方案:传入所有宗门作为 existed_sects (全集),虽然有点浪费,但不丢数据。
|
||||
# 更好的方案:修改 init_game,把 existed_sects 挂载到 world 上。
|
||||
|
||||
# 尝试从 world 属性获取(如果以后添加了)
|
||||
existed_sects = getattr(world, "existed_sects", [])
|
||||
if not existed_sects:
|
||||
|
||||
@@ -1,9 +1,7 @@
|
||||
"""
|
||||
配置管理模块
|
||||
使用OmegaConf读取config.yml和local_config.yml
|
||||
local_config.yml的优先级更高
|
||||
"""
|
||||
|
||||
from pathlib import Path
|
||||
from omegaconf import OmegaConf
|
||||
|
||||
@@ -34,6 +32,7 @@ def load_config():
|
||||
config = OmegaConf.merge(base_config, local_config)
|
||||
|
||||
# 把paths下的所有值pathlib化
|
||||
if hasattr(config, "paths"):
|
||||
for key, value in config.paths.items():
|
||||
config.paths[key] = Path(value)
|
||||
|
||||
|
||||
@@ -173,7 +173,7 @@ async def call_llm_with_task_name(
|
||||
return await call_llm_with_template(template_path, infos, mode, max_retries)
|
||||
|
||||
|
||||
def test_connectivity(mode: LLMMode = LLMMode.NORMAL, config: Optional[LLMConfig] = None) -> bool:
|
||||
def test_connectivity(mode: LLMMode = LLMMode.NORMAL, config: Optional[LLMConfig] = None) -> tuple[bool, str]:
|
||||
"""
|
||||
测试 LLM 服务连通性 (同步版本)
|
||||
|
||||
@@ -182,7 +182,7 @@ def test_connectivity(mode: LLMMode = LLMMode.NORMAL, config: Optional[LLMConfig
|
||||
config: 直接使用该配置进行测试
|
||||
|
||||
Returns:
|
||||
bool: 连接成功返回 True,失败返回 False
|
||||
tuple[bool, str]: (是否成功, 错误信息),成功时错误信息为空字符串
|
||||
"""
|
||||
try:
|
||||
if config is None:
|
||||
@@ -199,7 +199,22 @@ def test_connectivity(mode: LLMMode = LLMMode.NORMAL, config: Optional[LLMConfig
|
||||
else:
|
||||
# 直接调用 requests 实现
|
||||
_call_with_requests(config, "test")
|
||||
return True
|
||||
return True, ""
|
||||
except Exception as e:
|
||||
print(f"Connectivity test failed: {e}")
|
||||
return False
|
||||
error_msg = str(e)
|
||||
print(f"Connectivity test failed: {error_msg}")
|
||||
|
||||
# 解析常见错误并提供友好提示
|
||||
if "401" in error_msg or "invalid_api_key" in error_msg or "Incorrect API key" in error_msg:
|
||||
return False, "API Key 无效,请检查您的密钥是否正确"
|
||||
elif "403" in error_msg or "Forbidden" in error_msg:
|
||||
return False, "访问被拒绝,请检查您的权限或配额"
|
||||
elif "404" in error_msg:
|
||||
return False, "服务地址不存在,请检查 Base URL 是否正确"
|
||||
elif "timeout" in error_msg.lower():
|
||||
return False, "连接超时,请检查网络连接或服务地址"
|
||||
elif "Connection" in error_msg or "connect" in error_msg.lower():
|
||||
return False, "无法连接到服务器,请检查 Base URL 和网络"
|
||||
else:
|
||||
# 返回原始错误信息
|
||||
return False, error_msg
|
||||
|
||||
@@ -2,13 +2,13 @@
|
||||
|
||||
from enum import Enum
|
||||
from dataclasses import dataclass
|
||||
import os
|
||||
|
||||
from src.utils.config import CONFIG
|
||||
|
||||
class LLMMode(str, Enum):
|
||||
"""LLM 调用模式"""
|
||||
NORMAL = "normal"
|
||||
FAST = "fast"
|
||||
DEFAULT = "default"
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
@@ -21,7 +21,7 @@ class LLMConfig:
|
||||
@classmethod
|
||||
def from_mode(cls, mode: LLMMode) -> 'LLMConfig':
|
||||
"""
|
||||
根据模式创建配置
|
||||
根据模式创建配置,从 CONFIG 读取
|
||||
|
||||
Args:
|
||||
mode: LLM 调用模式
|
||||
@@ -29,36 +29,46 @@ class LLMConfig:
|
||||
Returns:
|
||||
LLMConfig: 配置对象
|
||||
"""
|
||||
from src.utils.config import CONFIG
|
||||
# 从 CONFIG 读取配置
|
||||
api_key = getattr(CONFIG.llm, "key", "")
|
||||
base_url = getattr(CONFIG.llm, "base_url", "")
|
||||
|
||||
# 根据模式选择模型
|
||||
model_name = (
|
||||
CONFIG.llm.model_name if mode == LLMMode.NORMAL
|
||||
else CONFIG.llm.fast_model_name
|
||||
)
|
||||
|
||||
# API Key 优先从环境变量读取
|
||||
api_key = CONFIG.llm.key
|
||||
model_name = ""
|
||||
if mode == LLMMode.FAST:
|
||||
model_name = getattr(CONFIG.llm, "fast_model_name", "")
|
||||
else:
|
||||
# NORMAL or DEFAULT fallback
|
||||
model_name = getattr(CONFIG.llm, "model_name", "")
|
||||
|
||||
return cls(
|
||||
model_name=model_name,
|
||||
api_key=api_key,
|
||||
base_url=CONFIG.llm.base_url
|
||||
base_url=base_url
|
||||
)
|
||||
|
||||
|
||||
def get_task_mode(task_name: str) -> LLMMode:
|
||||
"""
|
||||
获取指定任务的 LLM 调用模式
|
||||
|
||||
Args:
|
||||
task_name: 任务名称 (配置在 llm.default_modes 下的 key)
|
||||
|
||||
Returns:
|
||||
LLMMode: 对应的模式,如果未配置则默认返回 NORMAL
|
||||
根据任务名称获取 LLM 模式
|
||||
"""
|
||||
from src.utils.config import CONFIG
|
||||
# 从 CONFIG 读取全局模式
|
||||
global_mode = getattr(CONFIG.llm, "mode", "default").lower()
|
||||
|
||||
# 获取配置的模式字符串,默认 normal
|
||||
mode_str = getattr(CONFIG.llm.default_modes, task_name, "normal")
|
||||
return LLMMode(mode_str)
|
||||
if global_mode == "normal":
|
||||
return LLMMode.NORMAL
|
||||
elif global_mode == "fast":
|
||||
return LLMMode.FAST
|
||||
|
||||
# Default 模式:根据 task_name 从细粒度配置中获取
|
||||
# 如果配置了 default_modes,则根据任务名称返回对应模式
|
||||
default_modes = getattr(CONFIG.llm, "default_modes", {})
|
||||
if default_modes and task_name in default_modes:
|
||||
task_mode = default_modes[task_name].lower()
|
||||
if task_mode == "fast":
|
||||
return LLMMode.FAST
|
||||
else:
|
||||
return LLMMode.NORMAL
|
||||
|
||||
# 如果没有配置,默认返回 NORMAL
|
||||
return LLMMode.NORMAL
|
||||
|
||||
@@ -2,13 +2,6 @@ meta:
|
||||
version: "1.0.9"
|
||||
|
||||
llm:
|
||||
# 目前默认用的是阿里的千问大模型 api。
|
||||
# 如果你想用别家的,需要对应修改下面的base_url为对应的模型。
|
||||
# 填入对应的密钥,并且修改model_name和fast_model_name为对应的模型。
|
||||
key: "你的密钥"
|
||||
base_url: "https://dashscope.aliyuncs.com/compatible-mode/v1"
|
||||
model_name: "openai/qwen-plus" # 聪明的模型,负责难的任务
|
||||
fast_model_name: "openai/qwen-flash" # 快速的模型,负责简单的任务
|
||||
default_modes:
|
||||
action_decision: "normal"
|
||||
long_term_objective: "normal"
|
||||
@@ -17,7 +10,6 @@ llm:
|
||||
relation_resolver: "fast"
|
||||
story_teller: "fast"
|
||||
interaction_feedback: "fast"
|
||||
mode: "default" # default: 使用default_modes中的模式,normal: 均使用normal模式,fast: 均使用fast模式
|
||||
|
||||
paths:
|
||||
templates: static/templates/
|
||||
|
||||
@@ -19,6 +19,7 @@ const socketStore = useSocketStore()
|
||||
|
||||
const showMenu = ref(false)
|
||||
const isManualPaused = ref(false)
|
||||
const menuDefaultTab = ref<'save' | 'load' | 'create' | 'delete' | 'llm'>('load')
|
||||
|
||||
onMounted(async () => {
|
||||
// 初始化 Socket 连接
|
||||
@@ -28,6 +29,15 @@ onMounted(async () => {
|
||||
window.addEventListener('keydown', handleKeydown)
|
||||
})
|
||||
|
||||
// 导出方法供 socket store 调用
|
||||
function openLLMConfig() {
|
||||
menuDefaultTab.value = 'llm'
|
||||
showMenu.value = true
|
||||
}
|
||||
|
||||
// 暴露给全局以便 socket store 可以调用
|
||||
;(window as any).__openLLMConfig = openLLMConfig
|
||||
|
||||
onUnmounted(() => {
|
||||
socketStore.disconnect()
|
||||
window.removeEventListener('keydown', handleKeydown)
|
||||
@@ -116,6 +126,7 @@ watch([showMenu, isManualPaused], ([menuVisible, manualPaused]) => {
|
||||
|
||||
<SystemMenu
|
||||
:visible="showMenu"
|
||||
:default-tab="menuDefaultTab"
|
||||
@close="handleMenuClose"
|
||||
/>
|
||||
</div>
|
||||
|
||||
@@ -8,11 +8,13 @@ const API_BASE = import.meta.env.VITE_API_TARGET || '';
|
||||
|
||||
export class ApiError extends Error {
|
||||
public status: number;
|
||||
public response: { data: any };
|
||||
|
||||
constructor(status: number, message: string) {
|
||||
constructor(status: number, message: string, responseData?: any) {
|
||||
super(message);
|
||||
this.status = status;
|
||||
this.name = 'ApiError';
|
||||
this.response = { data: responseData || {} };
|
||||
}
|
||||
}
|
||||
|
||||
@@ -21,7 +23,21 @@ async function request<T>(path: string, options: RequestInit = {}): Promise<T> {
|
||||
const response = await fetch(url, options);
|
||||
|
||||
if (!response.ok) {
|
||||
throw new ApiError(response.status, `Request failed: ${response.statusText}`);
|
||||
// 尝试解析错误响应的 JSON
|
||||
let errorData = null;
|
||||
let errorMessage = `Request failed: ${response.statusText}`;
|
||||
|
||||
try {
|
||||
errorData = await response.json();
|
||||
// 如果后端返回了 detail 字段,使用它作为错误消息
|
||||
if (errorData?.detail) {
|
||||
errorMessage = errorData.detail;
|
||||
}
|
||||
} catch {
|
||||
// 如果解析失败,使用默认错误消息
|
||||
}
|
||||
|
||||
throw new ApiError(response.status, errorMessage, errorData);
|
||||
}
|
||||
|
||||
// 假设后端总是返回 JSON
|
||||
|
||||
@@ -7,29 +7,36 @@ import LLMConfigPanel from './game/panels/system/LLMConfigPanel.vue'
|
||||
|
||||
const props = defineProps<{
|
||||
visible: boolean
|
||||
defaultTab?: 'save' | 'load' | 'create' | 'delete' | 'llm'
|
||||
}>()
|
||||
|
||||
const emit = defineEmits<{
|
||||
(e: 'close'): void
|
||||
}>()
|
||||
|
||||
const activeTab = ref<'save' | 'load' | 'create' | 'delete' | 'llm'>('load')
|
||||
const activeTab = ref<'save' | 'load' | 'create' | 'delete' | 'llm'>(props.defaultTab || 'load')
|
||||
|
||||
function switchTab(tab: typeof activeTab.value) {
|
||||
activeTab.value = tab
|
||||
}
|
||||
|
||||
// Reset tab when reopening
|
||||
// 监听 defaultTab 变化
|
||||
watch(() => props.defaultTab, (newTab) => {
|
||||
if (newTab) {
|
||||
activeTab.value = newTab
|
||||
}
|
||||
})
|
||||
|
||||
// 当菜单打开时,如果有 defaultTab 就使用它
|
||||
watch(() => props.visible, (val) => {
|
||||
if (val) {
|
||||
// Do not reset activeTab to keep user context, or reset if preferred
|
||||
// activeTab.value = 'load'
|
||||
if (val && props.defaultTab) {
|
||||
activeTab.value = props.defaultTab
|
||||
}
|
||||
})
|
||||
</script>
|
||||
|
||||
<template>
|
||||
<div v-if="visible" class="system-menu-overlay" @click.self="emit('close')">
|
||||
<div v-if="visible" class="system-menu-overlay">
|
||||
<div class="system-menu">
|
||||
<div class="menu-header">
|
||||
<h2>系统菜单</h2>
|
||||
|
||||
@@ -42,6 +42,37 @@ export const useSocketStore = defineStore('socket', () => {
|
||||
uiStore.refreshDetail();
|
||||
}
|
||||
}
|
||||
// ===== 处理 LLM 配置要求消息 =====
|
||||
else if (data.type === 'llm_config_required') {
|
||||
console.warn('LLM 配置要求:', data.error);
|
||||
|
||||
// 显示错误提示
|
||||
const message = data.error || 'LLM 连接失败,请配置';
|
||||
|
||||
// 通过全局方法打开 LLM 配置界面
|
||||
if ((window as any).__openLLMConfig) {
|
||||
(window as any).__openLLMConfig();
|
||||
}
|
||||
|
||||
// 可以选择在这里显示一个提示消息
|
||||
// 需要引入 message API 或者通过其他方式
|
||||
setTimeout(() => {
|
||||
alert(`${message}\n\n请在打开的配置界面中完成 LLM 设置。`);
|
||||
}, 500);
|
||||
}
|
||||
// ===== 处理游戏重新初始化消息 =====
|
||||
else if (data.type === 'game_reinitialized') {
|
||||
console.log('游戏重新初始化:', data.message);
|
||||
|
||||
// 刷新世界状态
|
||||
worldStore.initialize().catch(console.error);
|
||||
|
||||
// 显示成功提示
|
||||
setTimeout(() => {
|
||||
alert(data.message || 'LLM 配置成功,游戏已重新初始化');
|
||||
}, 300);
|
||||
}
|
||||
// ===== LLM 消息处理结束 =====
|
||||
});
|
||||
|
||||
// Connect socket
|
||||
|
||||
Reference in New Issue
Block a user