mirror of
https://github.com/zhenxun-org/zhenxun_bot.git
synced 2026-10-10 06:10:01 +08:00
✨ feat(llm): 新增LLM模型管理插件并增强API密钥管理 (#1972)
🔧 新增功能: - LLM模型管理插件 (builtin_plugins/llm_manager/) • llm list - 查看可用模型列表 (图片格式) • llm info - 查看模型详细信息 (Markdown图片) • llm default - 管理全局默认模型 • llm test - 测试模型连通性 • llm keys - 查看API Key状态 (表格图片，含健康度/成功率/延迟) • llm reset-key - 重置API Key失败状态 🏗️ 架构重构: - 会话管理: AI/AIConfig 类迁移至独立的 session.py - 类型定义: TaskType 枚举移至 types/enums.py - API增强: • chat() 函数返回完整 LLMResponse，支持工具调用 • 新增 generate() 函数用于一次性响应生成 • 统一API调用核心方法 _perform_api_call，返回使用的API密钥 🚀 密钥管理增强: - 详细状态跟踪: 健康度、成功率、平均延迟、错误信息、建议操作 - 状态持久化: 启动时加载，关闭时自动保存密钥状态 - 智能冷却策略: 根据错误类型设置不同冷却时间 - 延迟监控: with_smart_retry 记录API调用延迟并更新统计 Co-authored-by: webjoin111 <455457521@qq.com> Co-authored-by: HibiKier <45528451+HibiKier@users.noreply.github.com>
This commit is contained in:
Rumio
co-authored by
webjoin111
HibiKier
webjoin111
HibiKier
parent
8649aaaa54
commit
46a0768a45
@@ -0,0 +1,171 @@
|
||||
from nonebot.permission import SUPERUSER
|
||||
from nonebot.plugin import PluginMetadata
|
||||
from nonebot_plugin_alconna import (
|
||||
Alconna,
|
||||
Args,
|
||||
Arparma,
|
||||
Match,
|
||||
Option,
|
||||
Query,
|
||||
Subcommand,
|
||||
on_alconna,
|
||||
store_true,
|
||||
)
|
||||
|
||||
from zhenxun.configs.utils import PluginExtraData
|
||||
from zhenxun.services.log import logger
|
||||
from zhenxun.utils.enum import PluginType
|
||||
from zhenxun.utils.message import MessageUtils
|
||||
|
||||
from .data_source import DataSource
|
||||
from .presenters import Presenters
|
||||
|
||||
__plugin_meta__ = PluginMetadata(
|
||||
name="LLM模型管理",
|
||||
description="查看和管理大语言模型服务。",
|
||||
usage="""
|
||||
LLM模型管理 (SUPERUSER)
|
||||
|
||||
llm list [--all]
|
||||
- 查看可用模型列表。
|
||||
- --all: 显示包括不可用在内的所有模型。
|
||||
|
||||
llm info <Provider/ModelName>
|
||||
- 查看指定模型的详细信息和能力。
|
||||
|
||||
llm default [Provider/ModelName]
|
||||
- 查看或设置全局默认模型。
|
||||
- 不带参数: 查看当前默认模型。
|
||||
- 带参数: 设置新的默认模型。
|
||||
- 例子: llm default Gemini/gemini-2.0-flash
|
||||
|
||||
llm test <Provider/ModelName>
|
||||
- 测试指定模型的连通性和API Key有效性。
|
||||
|
||||
llm keys <ProviderName>
|
||||
- 查看指定提供商的所有API Key状态。
|
||||
|
||||
llm reset-key <ProviderName> [--key <api_key>]
|
||||
- 重置提供商的所有或指定API Key的失败状态。
|
||||
""",
|
||||
extra=PluginExtraData(
|
||||
author="HibiKier",
|
||||
version="1.0.0",
|
||||
plugin_type=PluginType.SUPERUSER,
|
||||
).to_dict(),
|
||||
)
|
||||
|
||||
llm_cmd = on_alconna(
|
||||
Alconna(
|
||||
"llm",
|
||||
Subcommand("list", alias=["ls"], help_text="查看模型列表"),
|
||||
Subcommand("info", Args["model_name", str], help_text="查看模型详情"),
|
||||
Subcommand("default", Args["model_name?", str], help_text="查看或设置默认模型"),
|
||||
Subcommand(
|
||||
"test", Args["model_name", str], alias=["ping"], help_text="测试模型连通性"
|
||||
),
|
||||
Subcommand("keys", Args["provider_name", str], help_text="查看API密钥状态"),
|
||||
Subcommand(
|
||||
"reset-key",
|
||||
Args["provider_name", str],
|
||||
Option("--key", Args["api_key", str], help_text="指定要重置的API Key"),
|
||||
help_text="重置API Key状态",
|
||||
),
|
||||
Option("--all", action=store_true, help_text="显示所有条目"),
|
||||
),
|
||||
permission=SUPERUSER,
|
||||
priority=5,
|
||||
block=True,
|
||||
)
|
||||
|
||||
|
||||
@llm_cmd.assign("list")
|
||||
async def handle_list(arp: Arparma, show_all: Query[bool] = Query("all")):
|
||||
"""处理 'llm list' 命令"""
|
||||
logger.info("获取LLM模型列表", command="LLM Manage", session=arp.header_result)
|
||||
models = await DataSource.get_model_list(show_all=show_all.result)
|
||||
|
||||
image = await Presenters.format_model_list_as_image(models, show_all.result)
|
||||
await llm_cmd.finish(MessageUtils.build_message(image))
|
||||
|
||||
|
||||
@llm_cmd.assign("info")
|
||||
async def handle_info(arp: Arparma, model_name: Match[str]):
|
||||
"""处理 'llm info' 命令"""
|
||||
logger.info(
|
||||
f"获取模型详情: {model_name.result}",
|
||||
command="LLM Manage",
|
||||
session=arp.header_result,
|
||||
)
|
||||
details = await DataSource.get_model_details(model_name.result)
|
||||
if not details:
|
||||
await llm_cmd.finish(f"未找到模型: {model_name.result}")
|
||||
|
||||
image_bytes = await Presenters.format_model_details_as_markdown_image(details)
|
||||
await llm_cmd.finish(MessageUtils.build_message(image_bytes))
|
||||
|
||||
|
||||
@llm_cmd.assign("default")
|
||||
async def handle_default(arp: Arparma, model_name: Match[str]):
|
||||
"""处理 'llm default' 命令"""
|
||||
if model_name.available:
|
||||
logger.info(
|
||||
f"设置默认模型为: {model_name.result}",
|
||||
command="LLM Manage",
|
||||
session=arp.header_result,
|
||||
)
|
||||
success, message = await DataSource.set_default_model(model_name.result)
|
||||
await llm_cmd.finish(message)
|
||||
else:
|
||||
logger.info("查看默认模型", command="LLM Manage", session=arp.header_result)
|
||||
current_default = await DataSource.get_default_model()
|
||||
await llm_cmd.finish(f"当前全局默认模型为: {current_default or '未设置'}")
|
||||
|
||||
|
||||
@llm_cmd.assign("test")
|
||||
async def handle_test(arp: Arparma, model_name: Match[str]):
|
||||
"""处理 'llm test' 命令"""
|
||||
logger.info(
|
||||
f"测试模型连通性: {model_name.result}",
|
||||
command="LLM Manage",
|
||||
session=arp.header_result,
|
||||
)
|
||||
await llm_cmd.send(f"正在测试模型 '{model_name.result}'，请稍候...")
|
||||
|
||||
success, message = await DataSource.test_model_connectivity(model_name.result)
|
||||
await llm_cmd.finish(message)
|
||||
|
||||
|
||||
@llm_cmd.assign("keys")
|
||||
async def handle_keys(arp: Arparma, provider_name: Match[str]):
|
||||
"""处理 'llm keys' 命令"""
|
||||
logger.info(
|
||||
f"查看提供商API Key状态: {provider_name.result}",
|
||||
command="LLM Manage",
|
||||
session=arp.header_result,
|
||||
)
|
||||
sorted_stats = await DataSource.get_key_status(provider_name.result)
|
||||
if not sorted_stats:
|
||||
await llm_cmd.finish(
|
||||
f"未找到提供商 '{provider_name.result}' 或其没有配置API Keys。"
|
||||
)
|
||||
|
||||
image = await Presenters.format_key_status_as_image(
|
||||
provider_name.result, sorted_stats
|
||||
)
|
||||
await llm_cmd.finish(MessageUtils.build_message(image))
|
||||
|
||||
|
||||
@llm_cmd.assign("reset-key")
|
||||
async def handle_reset_key(
|
||||
arp: Arparma, provider_name: Match[str], api_key: Match[str]
|
||||
):
|
||||
"""处理 'llm reset-key' 命令"""
|
||||
key_to_reset = api_key.result if api_key.available else None
|
||||
log_msg = f"重置 {provider_name.result} 的 " + (
|
||||
"指定API Key" if key_to_reset else "所有API Keys"
|
||||
)
|
||||
logger.info(log_msg, command="LLM Manage", session=arp.header_result)
|
||||
|
||||
success, message = await DataSource.reset_key(provider_name.result, key_to_reset)
|
||||
await llm_cmd.finish(message)
|
||||
@@ -0,0 +1,120 @@
|
||||
import time
|
||||
from typing import Any
|
||||
|
||||
from zhenxun.services.llm import (
|
||||
LLMException,
|
||||
get_global_default_model_name,
|
||||
get_model_instance,
|
||||
list_available_models,
|
||||
set_global_default_model_name,
|
||||
)
|
||||
from zhenxun.services.llm.core import KeyStatus
|
||||
from zhenxun.services.llm.manager import (
|
||||
reset_key_status,
|
||||
)
|
||||
|
||||
|
||||
class DataSource:
|
||||
"""LLM管理插件的数据源和业务逻辑"""
|
||||
|
||||
@staticmethod
|
||||
async def get_model_list(show_all: bool = False) -> list[dict[str, Any]]:
|
||||
"""获取模型列表"""
|
||||
models = list_available_models()
|
||||
if show_all:
|
||||
return models
|
||||
return [m for m in models if m.get("is_available", True)]
|
||||
|
||||
@staticmethod
|
||||
async def get_model_details(model_name_str: str) -> dict[str, Any] | None:
|
||||
"""获取指定模型的详细信息"""
|
||||
try:
|
||||
model = await get_model_instance(model_name_str)
|
||||
return {
|
||||
"provider_config": model.provider_config,
|
||||
"model_detail": model.model_detail,
|
||||
"capabilities": model.capabilities,
|
||||
}
|
||||
except LLMException:
|
||||
return None
|
||||
|
||||
@staticmethod
|
||||
async def get_default_model() -> str | None:
|
||||
"""获取全局默认模型"""
|
||||
return get_global_default_model_name()
|
||||
|
||||
@staticmethod
|
||||
async def set_default_model(model_name_str: str) -> tuple[bool, str]:
|
||||
"""设置全局默认模型"""
|
||||
success = set_global_default_model_name(model_name_str)
|
||||
if success:
|
||||
return True, f"✅ 成功将默认模型设置为: {model_name_str}"
|
||||
else:
|
||||
return False, f"❌ 设置失败，模型 '{model_name_str}' 不存在或无效。"
|
||||
|
||||
@staticmethod
|
||||
async def test_model_connectivity(model_name_str: str) -> tuple[bool, str]:
|
||||
"""测试模型连通性"""
|
||||
start_time = time.monotonic()
|
||||
try:
|
||||
async with await get_model_instance(model_name_str) as model:
|
||||
await model.generate_text("你好")
|
||||
end_time = time.monotonic()
|
||||
latency = (end_time - start_time) * 1000
|
||||
return (
|
||||
True,
|
||||
f"✅ 模型 '{model_name_str}' 连接成功！\n响应延迟: {latency:.2f} ms",
|
||||
)
|
||||
except LLMException as e:
|
||||
return (
|
||||
False,
|
||||
f"❌ 模型 '{model_name_str}' 连接测试失败:\n"
|
||||
f"{e.user_friendly_message}\n错误码: {e.code.name}",
|
||||
)
|
||||
except Exception as e:
|
||||
return False, f"❌ 测试时发生未知错误: {e!s}"
|
||||
|
||||
@staticmethod
|
||||
async def get_key_status(provider_name: str) -> list[dict[str, Any]] | None:
|
||||
"""获取并排序指定提供商的API Key状态"""
|
||||
from zhenxun.services.llm.manager import get_key_usage_stats
|
||||
|
||||
all_stats = await get_key_usage_stats()
|
||||
provider_stats = all_stats.get(provider_name)
|
||||
|
||||
if not provider_stats or not provider_stats.get("key_stats"):
|
||||
return None
|
||||
|
||||
key_stats_dict = provider_stats["key_stats"]
|
||||
|
||||
stats_list = [
|
||||
{"key_id": key_id, **stats} for key_id, stats in key_stats_dict.items()
|
||||
]
|
||||
|
||||
def sort_key(item: dict[str, Any]):
|
||||
status_priority = item.get("status_enum", KeyStatus.UNUSED).value
|
||||
return (
|
||||
status_priority,
|
||||
100 - item.get("success_rate", 100.0),
|
||||
-item.get("total_calls", 0),
|
||||
)
|
||||
|
||||
sorted_stats_list = sorted(stats_list, key=sort_key)
|
||||
|
||||
return sorted_stats_list
|
||||
|
||||
@staticmethod
|
||||
async def reset_key(provider_name: str, api_key: str | None) -> tuple[bool, str]:
|
||||
"""重置API Key状态"""
|
||||
success = await reset_key_status(provider_name, api_key)
|
||||
if success:
|
||||
if api_key:
|
||||
if len(api_key) > 8:
|
||||
target = f"API Key '{api_key[:4]}...{api_key[-4:]}'"
|
||||
else:
|
||||
target = f"API Key '{api_key}'"
|
||||
else:
|
||||
target = "所有API Keys"
|
||||
return True, f"✅ 成功重置提供商 '{provider_name}' 的 {target} 的状态。"
|
||||
else:
|
||||
return False, "❌ 重置失败，请检查提供商名称或API Key是否正确。"
|
||||
@@ -0,0 +1,204 @@
|
||||
from typing import Any
|
||||
|
||||
from zhenxun.services.llm.core import KeyStatus
|
||||
from zhenxun.services.llm.types import ModelModality
|
||||
from zhenxun.utils._build_image import BuildImage
|
||||
from zhenxun.utils._image_template import ImageTemplate, Markdown, RowStyle
|
||||
|
||||
|
||||
def _format_seconds(seconds: int) -> str:
|
||||
"""将秒数格式化为 'Xm Ys' 或 'Xh Ym' 的形式"""
|
||||
if seconds <= 0:
|
||||
return "0s"
|
||||
if seconds < 60:
|
||||
return f"{seconds}s"
|
||||
|
||||
minutes, seconds = divmod(seconds, 60)
|
||||
if minutes < 60:
|
||||
return f"{minutes}m {seconds}s"
|
||||
|
||||
hours, minutes = divmod(minutes, 60)
|
||||
return f"{hours}h {minutes}m"
|
||||
|
||||
|
||||
class Presenters:
|
||||
"""格式化LLM管理插件的输出 (图片格式)"""
|
||||
|
||||
@staticmethod
|
||||
async def format_model_list_as_image(
|
||||
models: list[dict[str, Any]], show_all: bool
|
||||
) -> BuildImage:
|
||||
"""将模型列表格式化为表格图片"""
|
||||
title = "📋 LLM模型列表" + (" (所有已配置模型)" if show_all else " (仅可用)")
|
||||
|
||||
if not models:
|
||||
return await BuildImage.build_text_image(
|
||||
f"{title}\n\n当前没有配置任何LLM模型。"
|
||||
)
|
||||
|
||||
column_name = ["提供商", "模型名称", "API类型", "状态"]
|
||||
data_list = []
|
||||
for model in models:
|
||||
status_text = "✅ 可用" if model.get("is_available", True) else "❌ 不可用"
|
||||
embed_tag = " (Embed)" if model.get("is_embedding_model", False) else ""
|
||||
data_list.append(
|
||||
[
|
||||
model.get("provider_name", "N/A"),
|
||||
f"{model.get('model_name', 'N/A')}{embed_tag}",
|
||||
model.get("api_type", "N/A"),
|
||||
status_text,
|
||||
]
|
||||
)
|
||||
|
||||
return await ImageTemplate.table_page(
|
||||
head_text=title,
|
||||
tip_text="使用 `llm info <Provider/ModelName>` 查看详情",
|
||||
column_name=column_name,
|
||||
data_list=data_list,
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
async def format_model_details_as_markdown_image(details: dict[str, Any]) -> bytes:
|
||||
"""将模型详情格式化为Markdown图片"""
|
||||
provider = details["provider_config"]
|
||||
model = details["model_detail"]
|
||||
caps = details["capabilities"]
|
||||
|
||||
cap_list = []
|
||||
if ModelModality.IMAGE in caps.input_modalities:
|
||||
cap_list.append("视觉")
|
||||
if ModelModality.VIDEO in caps.input_modalities:
|
||||
cap_list.append("视频")
|
||||
if ModelModality.AUDIO in caps.input_modalities:
|
||||
cap_list.append("音频")
|
||||
if caps.supports_tool_calling:
|
||||
cap_list.append("工具调用")
|
||||
if caps.is_embedding_model:
|
||||
cap_list.append("文本嵌入")
|
||||
|
||||
md = Markdown()
|
||||
md.head(f"🔎 模型详情: {provider.name}/{model.model_name}", level=1)
|
||||
md.text("---")
|
||||
md.head("提供商信息", level=2)
|
||||
md.list(
|
||||
[
|
||||
f"**名称**: {provider.name}",
|
||||
f"**API 类型**: {provider.api_type}",
|
||||
f"**API Base**: {provider.api_base or '默认'}",
|
||||
]
|
||||
)
|
||||
md.head("模型详情", level=2)
|
||||
|
||||
temp_value = model.temperature or provider.temperature or "未设置"
|
||||
token_value = model.max_tokens or provider.max_tokens or "未设置"
|
||||
|
||||
md.list(
|
||||
[
|
||||
f"**名称**: {model.model_name}",
|
||||
f"**默认温度**: {temp_value}",
|
||||
f"**最大Token**: {token_value}",
|
||||
f"**核心能力**: {', '.join(cap_list) or '纯文本'}",
|
||||
]
|
||||
)
|
||||
|
||||
return await md.build()
|
||||
|
||||
@staticmethod
|
||||
async def format_key_status_as_image(
|
||||
provider_name: str, sorted_stats: list[dict[str, Any]]
|
||||
) -> BuildImage:
|
||||
"""将已排序的、详细的API Key状态格式化为表格图片"""
|
||||
title = f"🔑 '{provider_name}' API Key 状态"
|
||||
|
||||
if not sorted_stats:
|
||||
return await BuildImage.build_text_image(
|
||||
f"{title}\n\n该提供商没有配置API Keys。"
|
||||
)
|
||||
|
||||
def _status_row_style(column: str, text: str) -> RowStyle:
|
||||
style = RowStyle()
|
||||
if column == "状态":
|
||||
if "✅ 健康" in text:
|
||||
style.font_color = "#67C23A"
|
||||
elif "⚠️ 告警" in text:
|
||||
style.font_color = "#E6A23C"
|
||||
elif "❌ 错误" in text or "🚫" in text:
|
||||
style.font_color = "#F56C6C"
|
||||
elif "❄️ 冷却中" in text:
|
||||
style.font_color = "#409EFF"
|
||||
elif column == "成功率":
|
||||
try:
|
||||
if text != "N/A":
|
||||
rate = float(text.replace("%", ""))
|
||||
if rate < 80:
|
||||
style.font_color = "#F56C6C"
|
||||
elif rate < 95:
|
||||
style.font_color = "#E6A23C"
|
||||
except (ValueError, TypeError):
|
||||
pass
|
||||
return style
|
||||
|
||||
column_name = [
|
||||
"Key (部分)",
|
||||
"状态",
|
||||
"总调用",
|
||||
"成功率",
|
||||
"平均延迟(s)",
|
||||
"上次错误",
|
||||
"建议操作",
|
||||
]
|
||||
data_list = []
|
||||
|
||||
for key_info in sorted_stats:
|
||||
status_enum: KeyStatus = key_info["status_enum"]
|
||||
|
||||
if status_enum == KeyStatus.COOLDOWN:
|
||||
cooldown_seconds = int(key_info["cooldown_seconds_left"])
|
||||
formatted_time = _format_seconds(cooldown_seconds)
|
||||
status_text = f"❄️ 冷却中({formatted_time})"
|
||||
else:
|
||||
status_text = {
|
||||
KeyStatus.DISABLED: "🚫 永久禁用",
|
||||
KeyStatus.ERROR: "❌ 错误",
|
||||
KeyStatus.WARNING: "⚠️ 告警",
|
||||
KeyStatus.HEALTHY: "✅ 健康",
|
||||
KeyStatus.UNUSED: "⚪️ 未使用",
|
||||
}.get(status_enum, "❔ 未知")
|
||||
|
||||
total_calls = key_info["total_calls"]
|
||||
total_calls_text = (
|
||||
f"{key_info['success_count']}/{total_calls}"
|
||||
if total_calls > 0
|
||||
else "0/0"
|
||||
)
|
||||
|
||||
success_rate = key_info["success_rate"]
|
||||
success_rate_text = f"{success_rate:.1f}%" if total_calls > 0 else "N/A"
|
||||
|
||||
avg_latency = key_info["avg_latency"]
|
||||
avg_latency_text = f"{avg_latency / 1000:.2f}" if avg_latency > 0 else "N/A"
|
||||
|
||||
last_error = key_info.get("last_error") or "-"
|
||||
if len(last_error) > 25:
|
||||
last_error = last_error[:22] + "..."
|
||||
|
||||
data_list.append(
|
||||
[
|
||||
key_info["key_id"],
|
||||
status_text,
|
||||
total_calls_text,
|
||||
success_rate_text,
|
||||
avg_latency_text,
|
||||
last_error,
|
||||
key_info["suggested_action"],
|
||||
]
|
||||
)
|
||||
|
||||
return await ImageTemplate.table_page(
|
||||
head_text=title,
|
||||
tip_text="使用 `llm reset-key <Provider>` 重置Key状态",
|
||||
column_name=column_name,
|
||||
data_list=data_list,
|
||||
text_style=_status_row_style,
|
||||
column_space=15,
|
||||
)
|
||||
Reference in New Issue
Block a user