refactor: 单图方案重构 + 动态模型选择 + chat_services优化
All checks were successful
构建并部署 AI Agent 服务 / deploy (push) Successful in 12m9s

## 核心改动

### 1. 单图方案重构
- 删除了多图(self.graphs),改为单图(self.graph)
- 新增 MainGraphState.current_model 字段用于运行时注入模型
- llm_call 节点改为动态选择模型(create_dynamic_llm_call_node)

### 2. chat_services 优化
- 添加 _cached_services 缓存,避免重复初始化
- 新增 get_cached_chat_services() 函数,用于单图注入
- 新增 _check_http_service_available() 统一HTTP探测逻辑
- 减少重复代码,LocalVLLMChatProvider和LocalSmallModelProvider共用探测方法

### 3. AIAgentService 重构
- initialize() 只构建一次图,传入 chat_services 字典
- 新增 _resolve_model() 模型回退逻辑
- 新增 _build_invocation() 统一构建调用参数
- process_message() 和 process_message_stream() 改为注入 current_model
- 流式处理代码拆分,增加可读性

### 4. 新增和删除文件
- 新增:backend/app/main_graph/main_graph_builder.py(图构建)
- 新增:backend/app/main_graph/subgraph_wrapper.py(子图封装)
- 新增:tools/test/test_tavily_search.py(测试)
- 删除:backend/app/main_graph/graph.py(旧图)
- 删除:backend/app/main_graph/utils/main_graph_builder.py(旧构建器)
- 删除:backend/app/main_graph/utils/__init__.py

### 5. 其他更新
- README.md:新增模型服务使用情况详解章节
- backend/app/model_services/__init__.py:新增 get_cached_chat_services 导出

## 方案优势

- 内存优化:N张图 → 1张图
- 灵活性:运行时动态选择模型,支持同会话不同模型
- 性能:模型服务缓存,初始化仅一次
- 可维护性:减少重复代码,统一HTTP探测逻辑
This commit is contained in:
2026-05-05 17:30:55 +08:00
parent 8b5fbbd395
commit b5c15ef445
25 changed files with 1225 additions and 830 deletions

View File

@@ -6,8 +6,8 @@
from .reasoning import react_reason_node
from .web_search import web_search_node
from .error_handling import error_handling_node
from .routing import init_state_node, route_by_reasoning
from .llm_call import create_llm_call_node
from .routing import init_state_node, route_by_reasoning, should_summarize
from .llm_call import create_dynamic_llm_call_node
from .rag_nodes import rag_retrieve_node, rag_re_retrieve_node
# 记忆节点
@@ -38,7 +38,8 @@ __all__ = [
"web_search_node",
"error_handling_node",
"route_by_reasoning",
"create_llm_call_node",
"should_summarize",
"create_dynamic_llm_call_node",
"rag_retrieve_node",
"rag_re_retrieve_node",
# 记忆节点

View File

@@ -5,7 +5,7 @@ LLM 调用节点模块
import time
from typing import Any, Dict
from langchain_core.language_models import BaseLLM
from langchain_core.language_models import BaseChatModel
from langchain_core.messages import AIMessage
# 本地模块
@@ -14,29 +14,34 @@ from app.agent.prompts import create_system_prompt
from app.utils.logging import log_state_change
from app.logger import debug, info, error
def create_llm_call_node(llm, tools: list):
def create_dynamic_llm_call_node(chat_services: Dict[str, BaseChatModel], tools: list):
"""
工厂函数:创建 LLM 调用节点
工厂函数:创建动态 LLM 调用节点(根据 state.current_model 选择模型)
Args:
llm: LangChain LLM 实例
chat_services: 模型名称 -> ChatModel 实例 的字典
tools: 工具列表
Returns:
异步节点函数
"""
# 构建调用链
# 构建所有模型的 tools 绑定(避免每次调用都 bind
bound_models: Dict[str, Any] = {}
for name, llm in chat_services.items():
if tools:
bound_models[name] = llm.bind_tools(tools)
else:
bound_models[name] = llm
# 预构建 prompt
prompt = create_system_prompt(tools)
llm_with_tools = llm.bind_tools(tools)
# 恢复带 RunnableLambda 的链,并在下方使用 astream 遍历
chain = prompt | llm_with_tools
from langchain_core.runnables.config import RunnableConfig
async def call_llm(state: MainGraphState, config: RunnableConfig) -> Dict[str, Any]:
"""
LLM 调用节点(异步方法
LLM 调用节点(动态选择模型
Args:
state: 当前对话状态
@@ -46,7 +51,7 @@ def create_llm_call_node(llm, tools: list):
更新后的状态字典
"""
log_state_change("llm_call", state, "进入")
memory_context = getattr(state, "memory_context", "暂无用户信息")
start_time = time.time()
@@ -62,9 +67,20 @@ def create_llm_call_node(llm, tools: list):
"last_elapsed_time": elapsed_time,
"turns_since_last_summary": getattr(state, 'turns_since_last_summary', 0) + 1,
}
# 动态选择模型
model_name = getattr(state, "current_model", "")
if not model_name or model_name not in bound_models:
# 回退到第一个可用模型
fallback_name = next(iter(bound_models.keys()))
info(f"[llm_call] 模型 '{model_name}' 不可用,回退到 '{fallback_name}'")
model_name = fallback_name
llm_with_tools = bound_models[model_name]
info(f"[llm_call] 使用模型: {model_name}")
try:
# 添加 RAG 上下文到消息
# 添加上下文到消息
messages_with_context = list(state.messages)
if state.rag_context:
from langchain_core.messages import SystemMessage
@@ -77,9 +93,10 @@ def create_llm_call_node(llm, tools: list):
break
if not inserted:
messages_with_context.insert(0, rag_system_msg)
# 恢复为:手动进行 astream并将所有的 chunk 拼接成最终的 response 返回。
# LangGraph 会自动监听这期间产生的所有 token。
chain = prompt | llm_with_tools
chunks = []
async for chunk in chain.astream(
{
@@ -89,7 +106,7 @@ def create_llm_call_node(llm, tools: list):
config=config
):
chunks.append(chunk)
# 将所有 chunk 合并成最终的 AIMessage
if chunks:
response = chunks[0]
@@ -97,14 +114,14 @@ def create_llm_call_node(llm, tools: list):
response = response + chunk
else:
response = AIMessage(content="")
elapsed_time = time.time() - start_time
# 提取 token 用量(兼容不同 LLM 提供商的元数据格式)
token_usage = {}
input_tokens = 0
output_tokens = 0
# 尝试从 response_metadata 中提取
if hasattr(response, 'response_metadata') and response.response_metadata:
meta = response.response_metadata
@@ -112,33 +129,33 @@ def create_llm_call_node(llm, tools: list):
token_usage = meta['token_usage']
elif 'usage' in meta:
token_usage = meta['usage']
# 尝试从 additional_kwargs 中提取
if not token_usage and hasattr(response, 'additional_kwargs'):
add_kwargs = response.additional_kwargs
if 'llm_output' in add_kwargs and 'token_usage' in add_kwargs['llm_output']:
token_usage = add_kwargs['llm_output']['token_usage']
# 提取具体的 token 数值
if token_usage:
input_tokens = token_usage.get('prompt_tokens', token_usage.get('input_tokens', 0))
output_tokens = token_usage.get('completion_tokens', token_usage.get('output_tokens', 0))
# 打印 LLM 的完整输出
debug("\n" + "="*80)
debug("📥 [LLM输出] 模型返回的完整响应:")
debug(f"📥 [LLM输出] 模型: {model_name} 返回的完整响应:")
debug(f" 消息类型: {response.type.upper()}")
debug(f" 内容长度: {len(str(response.content))} 字符")
debug("-"*80)
debug(f"{response.content}")
# 打印响应统计信息
info(f"⏱️ [LLM统计] 调用耗时: {elapsed_time:.2f}")
info(f"📊 [LLM统计] Token用量: 输入={input_tokens}, 输出={output_tokens}, 总计={input_tokens + output_tokens}")
if token_usage:
debug(f"📋 [LLM统计] 详细用量: {token_usage}")
debug("="*80 + "\n")
# 检查是否有工具调用
has_tool_calls = hasattr(response, 'tool_calls') and len(response.tool_calls) > 0
@@ -151,21 +168,22 @@ def create_llm_call_node(llm, tools: list):
"final_result": response.content,
"success": True,
"current_phase": "done",
"has_tool_calls": has_tool_calls
"has_tool_calls": has_tool_calls,
"current_model": model_name # 记录实际使用的模型
}
log_state_change("llm_call", state, "离开")
return result
except Exception as e:
elapsed_time = time.time() - start_time
error(f"\n❌ [LLM错误] 调用失败 (耗时: {elapsed_time:.2f}秒)")
error(f"\n❌ [LLM错误] 模型 {model_name} 调用失败 (耗时: {elapsed_time:.2f}秒)")
error(f" 错误类型: {type(e).__name__}")
error(f" 错误信息: {str(e)}")
import traceback
traceback.print_exc()
error(f"📋 堆栈: {traceback.format_exc()}")
debug("="*80 + "\n")
# 返回一个友好的错误消息
error_response = AIMessage(
content="抱歉,模型暂时无法响应,可能是网络超时或服务繁忙,请稍后再试。"
@@ -178,10 +196,11 @@ def create_llm_call_node(llm, tools: list):
"turns_since_last_summary": getattr(state, 'turns_since_last_summary', 0) + 1,
"final_result": "抱歉,模型暂时无法响应,可能是网络超时或服务繁忙,请稍后再试。",
"success": False,
"current_phase": "done"
"current_phase": "done",
"current_model": model_name
}
log_state_change("llm_call", state, "离开(异常)")
return error_result
return call_llm
return call_llm

View File

@@ -118,3 +118,21 @@ def route_by_reasoning(state: MainGraphState) -> str:
info(f"[条件路由] 动作={latest_action}, 目标={target}")
return target
# ========== 完成阶段条件路由函数 ==========
def should_summarize(state: MainGraphState) -> str:
"""
检查是否需要总结对话(对话足够长时)
Args:
state: 当前图状态
Returns:
"summarize""finalize"
"""
if state.turns_since_last_summary >= 5: # 每5轮对话总结一次
return "summarize"
else:
return "finalize"