1f81bef8c6
🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
275 lines
12 KiB
Python
275 lines
12 KiB
Python
# Copyright 2023 The Qwen team, Alibaba Group. All rights reserved.
|
||
#
|
||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||
# you may not use this file except in compliance with the License.
|
||
# You may obtain a copy of the License at
|
||
#
|
||
# http://www.apache.org/licenses/LICENSE-2.0
|
||
#
|
||
# Unless required by applicable law or agreed to in writing, software
|
||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||
# See the License for the specific language governing permissions and
|
||
# limitations under the License.
|
||
|
||
import copy
|
||
import json
|
||
import os
|
||
import time
|
||
from typing import Dict, Iterator, List, Literal, Optional, Union
|
||
|
||
from qwen_agent.agents import Assistant
|
||
from qwen_agent.llm.schema import ASSISTANT, FUNCTION, Message
|
||
from qwen_agent.llm.oai import TextChatAtOAI
|
||
|
||
class ModifiedAssistant(Assistant):
|
||
"""
|
||
修改后的 Assistant 子类，改变循环判断逻辑：
|
||
- 原始逻辑：如果没有使用工具，立即退出循环
|
||
- 修改后逻辑：如果没有使用工具，调用模型判断回答是否完整，如果不完整则继续循环
|
||
"""
|
||
|
||
def _is_retryable_error(self, error: Exception) -> bool:
|
||
"""判断错误是否可重试
|
||
|
||
Args:
|
||
error: 异常对象
|
||
|
||
Returns:
|
||
bool: 是否可重试
|
||
"""
|
||
error_str = str(error).lower()
|
||
retryable_indicators = [
|
||
'502', '500', '503', '504', # HTTP错误代码
|
||
'internal server error', # 内部服务器错误
|
||
'timeout', # 超时
|
||
'connection', # 连接错误
|
||
'network', # 网络错误
|
||
'rate', # 速率限制和相关错误
|
||
'quota', # 配额限制
|
||
'service unavailable', # 服务不可用
|
||
'provider returned error', # Provider错误
|
||
'model service error', # 模型服务错误
|
||
'temporary', # 临时错误
|
||
'retry' # 明确提示重试
|
||
]
|
||
return any(indicator in error_str for indicator in retryable_indicators)
|
||
|
||
def _call_llm_with_retry(self, messages: List[Message], functions=None, extra_generate_cfg=None, max_retries: int = 5) -> Iterator:
|
||
"""带重试机制的LLM调用
|
||
|
||
Args:
|
||
messages: 消息列表
|
||
functions: 函数列表
|
||
extra_generate_cfg: 额外生成配置
|
||
max_retries: 最大重试次数
|
||
|
||
Returns:
|
||
LLM响应流
|
||
|
||
Raises:
|
||
Exception: 重试次数耗尽后重新抛出原始异常
|
||
"""
|
||
for attempt in range(max_retries):
|
||
try:
|
||
return self._call_llm(messages=messages, functions=functions, extra_generate_cfg=extra_generate_cfg)
|
||
except Exception as e:
|
||
# 检查是否为可重试的错误
|
||
if self._is_retryable_error(e) and attempt < max_retries - 1:
|
||
delay = 2 ** attempt # 指数退避: 1s, 2s, 4s
|
||
print(f"LLM调用失败 (尝试 {attempt + 1}/{max_retries})，{delay}秒后重试: {str(e)}")
|
||
time.sleep(delay)
|
||
continue
|
||
else:
|
||
# 不可重试的错误或已达到最大重试次数
|
||
if attempt > 0:
|
||
print(f"LLM调用重试失败，已达到最大重试次数 {max_retries}")
|
||
raise
|
||
|
||
def _run(self, messages: List[Message], lang: Literal['en', 'zh'] = 'en', **kwargs) -> Iterator[List[Message]]:
|
||
|
||
message_list = copy.deepcopy(messages)
|
||
response = []
|
||
|
||
# 保持原有的最大调用次数限制
|
||
total_num_llm_calls_available = self.MAX_LLM_CALL_PER_RUN if hasattr(self, 'MAX_LLM_CALL_PER_RUN') else 10
|
||
num_llm_calls_available = total_num_llm_calls_available
|
||
while num_llm_calls_available > 0:
|
||
num_llm_calls_available -= 1
|
||
extra_generate_cfg = {'lang': lang}
|
||
if kwargs.get('seed') is not None:
|
||
extra_generate_cfg['seed'] = kwargs['seed']
|
||
|
||
output_stream = self._call_llm_with_retry(messages=message_list,
|
||
functions=[func.function for func in self.function_map.values()],
|
||
extra_generate_cfg=extra_generate_cfg)
|
||
output: List[Message] = []
|
||
for output in output_stream:
|
||
if output:
|
||
yield response + output
|
||
|
||
if output:
|
||
response.extend(output)
|
||
message_list.extend(output)
|
||
|
||
# 处理工具调用
|
||
used_any_tool = False
|
||
for out in output:
|
||
use_tool, tool_name, tool_args, _ = self._detect_tool(out)
|
||
if use_tool:
|
||
tool_result = self._call_tool(tool_name, tool_args, messages=message_list, **kwargs)
|
||
fn_msg = Message(role=FUNCTION,
|
||
name=tool_name,
|
||
content=tool_result,
|
||
extra={'function_id': out.extra.get('function_id', '1')})
|
||
message_list.append(fn_msg)
|
||
response.append(fn_msg)
|
||
yield response
|
||
used_any_tool = True
|
||
|
||
# 如果使用了工具，继续循环
|
||
if used_any_tool:
|
||
continue
|
||
# 如果没有使用工具，无调用次数，已经执行了2次以上循环，退出循环
|
||
if num_llm_calls_available ==0 or total_num_llm_calls_available - num_llm_calls_available >=2:
|
||
break
|
||
# 如果没有使用工具，还有调用次数，并只执行了1次，就需要调用模型判断回答是否完整（修复部分case不执行工具调用就停止的问题）
|
||
if num_llm_calls_available > 0:
|
||
# 构建判断消息 - 使用英文系统提示词，并在用户提示中包含具体内容
|
||
user_question = messages[-1].content if messages[-1].content else str(messages[-1])
|
||
assistant_response = output[-1].content if hasattr(output[-1], 'content') and output[-1].content else str(output[-1])
|
||
|
||
judge_messages = [
|
||
Message(role='system', content='''You are a professional conversation completeness evaluator. Your task is to determine whether an AI assistant has provided a complete and sufficient answer to a user's question based on strict evaluation criteria.
|
||
|
||
## Evaluation Criteria
|
||
|
||
### Complete Answer Characteristics (mark as "complete" if ANY of these are met):
|
||
1. **Information Completeness**: Provides core information users need, including specific parameters, specifications, data, or detailed descriptions
|
||
2. **Problem Resolution**: Directly addresses the user's question with clear answers or solutions
|
||
3. **Recommendation Adequacy**: When providing product/service recommendations, includes key information (model, parameters, price, features, etc.)
|
||
4. **Conclusion Clarity**: Provides clear conclusions, advice, or summaries
|
||
|
||
### Incomplete Answer Characteristics (mark as "incomplete" if ANY of these are met):
|
||
1. **Information Gaps**: Obviously lacks key information needed to answer the question
|
||
2. **Vague Responses**: Uses evasive language and avoids specific questions
|
||
3. **Unfulfilled Promises**: Promises to provide more information but doesn't deliver
|
||
4. **Needs Follow-up**: Clearly indicates more information is needed to continue answering
|
||
|
||
## Special Evaluation Rules
|
||
- **Recommendation Questions**: Consider complete if specific recommendations with key information are provided
|
||
- **Technical Questions**: Consider complete if technical details and specific data are provided
|
||
- **Comparison Questions**: Consider complete if valuable comparative information is provided
|
||
- **Inquiry Questions**: Consider complete if best effort answer is given based on available information
|
||
|
||
## Output Format
|
||
Only output "complete" or "incomplete" with no additional text or explanation.
|
||
|
||
Adhere strictly to these standards for objective evaluation.'''),
|
||
Message(role='user', content=f'''Please evaluate whether the following assistant response is complete based on the evaluation criteria.
|
||
|
||
USER QUESTION:
|
||
{user_question}
|
||
|
||
ASSISTANT RESPONSE:
|
||
{assistant_response}
|
||
|
||
Based on the evaluation criteria, is this response complete? Only reply with "complete" or "incomplete".''')
|
||
]
|
||
# 调用模型进行判断（不使用工具）
|
||
judge_stream = self._call_llm_with_retry(messages=judge_messages, functions=[], extra_generate_cfg={'lang': 'en'})
|
||
judge_output = []
|
||
for judge_msg in judge_stream:
|
||
if judge_msg:
|
||
judge_output.extend(judge_msg)
|
||
|
||
# 分析判断结果 - 使用正则匹配检测 complete
|
||
is_complete = False
|
||
if judge_output:
|
||
judge_content = judge_output[-1].content if hasattr(judge_output[-1], 'content') else str(judge_output[-1])
|
||
judge_content = judge_content.lower().strip()
|
||
print(judge_content)
|
||
# 使用正则匹配检测 complete
|
||
import re
|
||
|
||
if re.search(r'incomplete', judge_content):
|
||
is_complete = False
|
||
else:
|
||
is_complete = True
|
||
# 如果回答完整，退出循环；否则继续
|
||
if is_complete:
|
||
break
|
||
|
||
yield response
|
||
|
||
|
||
# Utility functions
|
||
def read_system_prompt():
|
||
"""读取通用的无状态系统prompt"""
|
||
with open("./prompt/system_prompt_default.md", "r", encoding="utf-8") as f:
|
||
return f.read().strip()
|
||
|
||
|
||
def read_mcp_settings():
|
||
"""读取MCP工具配置"""
|
||
with open("./mcp/mcp_settings.json", "r") as f:
|
||
mcp_settings_json = json.load(f)
|
||
return mcp_settings_json
|
||
|
||
|
||
def update_agent_llm(agent, model_name: str, api_key: str = None, model_server: str = None, generate_cfg: Dict = None, system_prompt: str = None, mcp_settings: List[Dict] = None):
|
||
"""动态更新助手实例的LLM和配置，支持从接口传入参数"""
|
||
# 获取基础配置
|
||
llm_config = {
|
||
"model": model_name,
|
||
"api_key": api_key,
|
||
"model_server": model_server,
|
||
"generate_cfg": generate_cfg if generate_cfg else {}
|
||
}
|
||
|
||
# 创建LLM实例
|
||
llm_instance = TextChatAtOAI(llm_config)
|
||
|
||
# 动态设置LLM
|
||
agent.llm = llm_instance
|
||
|
||
# 更新系统消息（如果提供）
|
||
if system_prompt:
|
||
agent.system_message = system_prompt
|
||
|
||
# 更新MCP设置（如果提供）
|
||
if mcp_settings:
|
||
agent.function_list = mcp_settings
|
||
|
||
return agent
|
||
|
||
|
||
# 向后兼容：保持原有的初始化函数接口
|
||
def init_modified_agent_service_with_files(rag_cfg=None,
|
||
model_name="qwen3-next", api_key=None,
|
||
model_server=None, generate_cfg=None,
|
||
system_prompt=None, mcp=None):
|
||
"""创建支持预加载文件的修改版助手实例"""
|
||
system = system_prompt if system_prompt else read_system_prompt()
|
||
tools = mcp if mcp else read_mcp_settings()
|
||
|
||
llm_config = {
|
||
"model": model_name,
|
||
"api_key": api_key,
|
||
"model_server": model_server,
|
||
"generate_cfg": generate_cfg if generate_cfg else {}
|
||
}
|
||
|
||
# 创建LLM实例
|
||
llm_instance = TextChatAtOAI(llm_config)
|
||
|
||
bot = ModifiedAssistant(
|
||
llm=llm_instance,
|
||
name="修改版数据检索助手",
|
||
description="基于智能判断循环终止的助手",
|
||
system_message=system,
|
||
function_list=tools,
|
||
)
|
||
|
||
return bot
|