import random from pathlib import Path import time import sys import os sys.path.append("./") from tools.yaml_operator import read_yaml from tools.log import CustomLogger import requests import queue current_file_path = os.path.abspath(__file__) Language_Path = os.path.dirname(os.path.dirname(os.path.dirname(current_file_path))) # print("Language_Path:",Language_Path) # 找到上三级目录的父级（project_root），然后拼接目标文件夹路径 MassageRobot_Dobot_Path = os.path.dirname(Language_Path) # print("MassageRobot_Dobot_Path:",MassageRobot_Dobot_Path) # 添加目标文件夹到系统路径 sys.path.append(MassageRobot_Dobot_Path) # # 测试 try: from LLM_dify import DifyClient except: # 外部调用 from .LLM_dify import DifyClient from datetime import datetime import json from openai import OpenAI import re import threading import subprocess import os import requests from VortXDB.client import VTXClient import asyncio import websockets class DashscopeClient: def __init__(self): # 获取密钥 self.weather_info=None self.vtxdb = VTXClient() self.Ali_OpenAI_api_key = self.vtxdb.get("robot_config", "Language.LLM.Ali_OpenAI_api_key") # print("self.Sillcon_OpenAI_api_key:",self.Sillcon_OpenAI_api_key) self.Ali_OpenAI_BaseUrl = self.vtxdb.get("robot_config", "Language.LLM.Ali_OpenAI_BaseUrl") self.client = OpenAI(api_key=self.Ali_OpenAI_api_key, base_url=self.Ali_OpenAI_BaseUrl) self.client_deep=OpenAI(api_key=self.vtxdb.get("robot_config", "Language.LLM.Sillcon_OpenAI_api_key"),base_url=self.vtxdb.get("robot_config", "Language.LLM.Sillcon_OpenAI_BaseUrl")) # self.client_search=DifyClient(base_url='http://124.71.62.243/v1',api_key='app-9vrC0QkaVFbj1g2rKiwkyzv3') # self.client_search_deep=DifyClient(base_url='http://124.71.62.243/v1',api_key='app-KFatoFKPMwjIt2za2paXvVA7') # self.massage_method_suggestion=DifyClient(base_url='http://124.71.62.243/v1',api_key='app-Kcm2KEaWmAIS5FiMmsX5hcyd') self.client_search=DifyClient(base_url='https://robotstorm.tech/dify/v1',api_key='app-9vrC0QkaVFbj1g2rKiwkyzv3') self.client_search_deep=DifyClient(base_url='https://robotstorm.tech/dify/v1',api_key='app-KFatoFKPMwjIt2za2paXvVA7') self.massage_method_suggestion=DifyClient(base_url='https://robotstorm.tech/dify/v1',api_key='app-WrJeh2wGNVcKoJtZPEK2EC27') # self.massage_method_suggestion=DifyClient(base_url='http://124.71.62.243/v1',api_key='app-RzktoHvGfosZmrWW5hwQyT5z') self.logger = CustomLogger() self.massage_status_languge={'task_time':'3','progress':'30','force': 0, 'press': 0, 'frequency': 0, 'temperature': 0, 'gear': 0,'body_part':'back','current_head':'thermotherapy_head','shake':5,'speed':1,'direction':1,'start_pos':'滑肉左','end_pos':'滑肉右','massage_path':'line'} self.deep_thinking_flag=False self.search_flag=False self.suggestion_flag=False self.suggestion_deekseek_flag=False self.stop_event = threading.Event() self.thread = None self.suggestion_mode_flag=False self.correct_flag=False # self.massage_status_languge=None # 初始化天气和新闻 # self.weather_and_news=Weather_AND_News(api_keys) # 读取prompt with open('LLM/config/classify_prompts_new.txt', 'r') as f: self.classify_prompts = f.read().strip() self.classify_message = [{'role': 'system', 'content': self.classify_prompts}] with open('LLM/config/auto_correction_prompts.txt', 'r') as f: self.chat_prompts1 = f.read().strip() self.chat_message1 = [{'role': 'system', 'content': self.chat_prompts1}] with open('LLM/config/classify_prompts_new.txt', 'r') as f: self.classify_prompts2 = f.read().strip() self.classify_message2 = [{'role': 'system', 'content': self.classify_prompts2}] with open('LLM/config/chat_prompts_new.txt', 'r') as f: self.chat_prompts = f.read().strip() self.chat_message = [{'role': 'system', 'content': self.chat_prompts}] self.weather_message = self.chat_message with open('LLM/config/retrieve_prompts.txt', 'r') as f: self.retrieve_prompts = f.read().strip() self.retrieve_message = [{'role': 'system', 'content': self.retrieve_prompts}] with open('LLM/config/adjust_volumn_prompts.txt', 'r') as f: self.adjust_volumn_prompts = f.read().strip() self.adjust_volumn_message = [{'role': 'system', 'content': self.adjust_volumn_prompts}] # with open('LLM/config/trajectory_retrieval_prompt.txt', 'r') as f: # self.trajectory_retrieval_prompt = f.read().strip() # self.trajectory_retrieval_message = [{'role': 'system', 'content': self.trajectory_retrieval_prompt}] with open('LLM/config/trajectory_config/message_retrieval_prompt.txt', 'r') as f: self.message_retrieval_prompt = f.read().strip() self.message_retrieval_message = [{'role': 'system', 'content': self.message_retrieval_prompt}] with open('LLM/config/trajectory_config/body_part_prompts.txt', 'r') as f: self.body_part_prompts = f.read().strip() self.body_part_message = [{'role': 'system', 'content': self.body_part_prompts}] with open('LLM/config/trajectory_config/reason.txt', 'r') as f: self.reason_prompts = f.read().strip() self.reason_message = [{'role': 'system', 'content': self.reason_prompts}] with open('LLM/config/trajectory_config/shoulder_limit_prompts.txt', 'r') as f: self.shoulder_limit_prompts = f.read().strip() with open('LLM/config/trajectory_config/back_limit_prompts.txt', 'r') as f: self.back_limit_prompts = f.read().strip() with open('LLM/config/trajectory_config/waist_limit_prompts.txt', 'r') as f: self.waist_limit_prompts = f.read().strip() with open('LLM/config/trajectory_config/belly_limit_prompts.txt', 'r') as f: self.belly_limit_prompts = f.read().strip() with open('LLM/config/trajectory_config/leg_limit_prompts.txt', 'r') as f: self.leg_limit_prompts = f.read().strip() self.model = { 'classify_model2': 'qwen2.5-72b-instruct', 'classify_model1': 'qwen2-7b-instruct', 'chat_model': 'qwen2-7b-instruct', 'chat_model1': 'qwen2-7b-instruct', 'retrieve_model': 'qwen2-7b-instruct', 'adjust_volumn_model': 'qwen2-7b-instruct', 'deep_thinking_model': 'deepseek-ai/DeepSeek-R1-Distill-Qwen-32B' } # self.location = self.get_current_position()[3:] # 定义操作字典 self.operations = { 'ksam': '按摩过程调整：开始,无', 'tzam': '按摩过程调整：停止,无', 'ldzd': '按摩过程调整：变重,一点', 'ldjx': '按摩过程调整：变轻,一点', 'wdzj': '按摩过程调整：升温,一点', 'wdjx': '按摩过程调整：降温,一点', 'dlzd': '按摩过程调整：增电,一点', 'dljx': '按摩过程调整：减电,一点', 'pljk': '按摩过程调整：增频,一点', 'pljd': '按摩过程调整：减频,一点', 'cjlz': '按摩过程调整：增冲,一点', 'cjlj': '按摩过程调整：减冲,一点', 'zszj': '按摩过程调整：增速,一点', 'zsjx': '按摩过程调整：减速,一点', 'gbzx': '按摩过程调整：换向,一点', 'tsgd': '按摩过程调整：升高,一点', 'jdgd': '按摩过程调整：降高,一点', 'dayh': '按摩过程调整：时间,加长', 'tgdw': '按摩过程调整：时间,跳过', 'jtrq': '', 'xzsj': '', 'jttq': 'weather', 'jtxw': 'news', 'qthd': '', 'bfyy': '音乐播放调整：播放,无', 'tzbf': '音乐播放调整：停播,无', 'ssyy': '音乐播放调整：上首', 'xsyy': '音乐播放调整：下首', } self.operations1 = { 'ksam': '按摩过程调整：开始,无', 'tzam': '按摩过程调整：停止,无', 'ldzd': '按摩过程调整：变重,较大幅度', 'ldjx': '按摩过程调整：变轻,较大幅度', 'wdzj': '按摩过程调整：升温,较大幅度', 'wdjx': '按摩过程调整：降温,较大幅度', 'dlzd': '按摩过程调整：增电,较大幅度', 'dljx': '按摩过程调整：减电,较大幅度', 'pljk': '按摩过程调整：增频,较大幅度', 'pljd': '按摩过程调整：减频,较大幅度', 'cjlz': '按摩过程调整：增冲,较大幅度', 'cjlj': '按摩过程调整：减冲,较大幅度', 'zszj': '按摩过程调整：增速,较大幅度', 'zsjx': '按摩过程调整：减速,较大幅度', 'gbzx': '按摩过程调整：换向,较大幅度', 'dayh': '按摩过程调整：时间,加长', 'tgdw': '按摩过程调整：时间,跳过', 'jtrq': '', 'xzsj': '', 'jttq': 'weather', 'jtxw': 'news', 'qthd': '', 'bfyy': '音乐播放调整：播放,无', 'tzbf': '音乐播放调整：停播,无', 'ssyy': '音乐播放调整：上首', 'xsyy': '音乐播放调整：下首', } self.task_mapping = { "finger": "指疗通络", "shockwave": "点阵按摩", "roller": "滚滚刺疗", "thermotherapy": "深部热疗", "stone": "温砭舒揉", "ball": "全能滚珠" } self.body_mapping = { "back": "背部", "belly": "腹部", "waist": "腰部", "shoulder": "肩颈", "leg": "腿部" } self.pathTypeMap = { 'line': "循经直推法", 'in_spiral': "螺旋内揉法", 'out_spiral': "螺旋外散法", 'ellipse': "周天环摩法", 'lemniscate': "双环疏经法", 'cycloid': "摆浪通络法", 'point': "定穴点按法", 'point_rub': "定点揉摩法" } # 获取位置信息 # def get_current_position(self, **kwargs): # """ 调用API获取位置 """ # try: # location = get_location()['城市'] # return '位置：' + location # except Exception as e: # self.logger.log_error("获取位置信息失败") # 获取时间信息 def xzsj(self, **kwargs): time_result = datetime.now().strftime('%Y-%m-%d %H:%M:%S') time_result = time_result.split(' ')[1] return '当前时间：'+time_result # 获取日期信息 def jtrq(self, **kwargs): time_result = datetime.now().strftime('%Y-%m-%d %H:%M:%S') date_result = time_result.split(' ')[0] print('当前日期：'+date_result) return '当前日期：'+date_result def get_volume(self): """ 查询当前音量并返回结果。 """ try: # 执行命令并捕获输出 result = subprocess.run( ["/usr/bin/amixer", "-D", "pulse", "get", "Master"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True ) if result.returncode == 0: # 解析结果 output = result.stdout return output else: print("查询音量失败:", result.stderr) return None except Exception as e: print("发生错误:", e) return None # 通用处理方法 def handle_request(self, operation, **kwargs): self.stop_thread() # 默认获取用户输入 human_input = kwargs.get('human_input') # 获取当前机器人状态值 function_to_status = { 'lddx': str(self.massage_status_languge['force']), 'wddx': str(self.massage_status_languge['temperature']), 'dldw': str(self.massage_status_languge['gear']), # 'pldw': str(self.massage_status_languge['frequency']), 'pldw': str(self.massage_status_languge['frequency']) if str(self.massage_status_languge['current_head']) == 'shockwave_head' else str(self.massage_status_languge['shake']) if str(self.massage_status_languge['current_head']) == 'thermotherapy_head' else str(self.massage_status_languge['frequency']), # 默认值 'cjdx': str(self.massage_status_languge['press']), 'amjd': str(self.massage_status_languge['progress']), 'amdt': str(self.massage_status_languge['current_head']), 'ambw': str(self.massage_status_languge['body_part']), 'gnjs': str(self.massage_status_languge['current_head']), 'zsdw': str(self.massage_status_languge['speed']) } status_result = function_to_status.get(operation, None) print('status_result:',status_result) if operation in ['lddx', 'wddx', 'dldw', 'pldw', 'cjdx','zsdw']: if operation in ['lddx', 'cjdx']: self.unit_status = "牛" elif operation=='wddx': self.unit_status = "档" elif operation in ['dldw', 'pldw', 'zsdw']: self.unit_status="档" human_input = f"{human_input} {status_result}{self.unit_status} " if operation == 'ambw': if status_result == 'back': human_input = f"{human_input} 目前按摩部位:{'背部'}" elif status_result == 'shoulder': human_input = f"{human_input} 目前按摩部位:{'肩颈'}" elif status_result == 'belly': human_input = f"{human_input} 目前按摩部位:{'腹部'}" elif status_result == 'back_shoulder': human_input = f"{human_input} 目前按摩部位:{'肩颈和背部'}" if operation in ['amdt','gnjs']: if status_result == 'thermotherapy_head': human_input = f"{human_input} 目前使用按摩头:{'热疗按摩头'}" elif status_result == 'shockwave_head': human_input = f"{human_input} 目前使用按摩头:{'冲击波按摩头'}" elif status_result == 'ball_head': human_input = f"{human_input} 目前使用按摩头:{'全能滚珠按摩头'}" elif status_result == 'finger_head': human_input = f"{human_input} 目前使用按摩头:{'指疗通络按摩头'}" elif status_result == 'roller_head': human_input = f"{human_input} 目前使用按摩头:{'狼牙滚珠按摩头'}" if operation == 'amjd': massage_progress=int(self.massage_status_languge['progress']) massage_task_time=int(self.massage_status_languge['task_time']) remaining_time = massage_task_time*12-(round((massage_task_time * 12) * massage_progress / 100)) print("massage_progress:",massage_progress) print("massage_task_time:",massage_task_time) print("remaining_time:",remaining_time) human_input=f"{human_input} 目前按摩进度:{massage_progress},按摩剩余时间:{remaining_time}" # 如果操作为 jtrq 或 xzsj，则获取相应函数的返回值作为 human_input if operation == 'jtrq': human_input = self.jtrq() elif operation == 'xzsj': human_input = self.xzsj() if operation == 'cqjm': try: # 提供密码通过标准输入传递给 sudo result = subprocess.run( ['/bin/sudo', '-S', 'systemctl', 'restart', 'ui_next_app'], input='jsfb\n', text=True, check=True ) print("服务 ui_next_app 已成功重启") except subprocess.CalledProcessError as e: print(f"重启服务失败，错误信息: {e}") if operation == 'yltj': try: self.adjust_volumn_message = [{'role': 'system', 'content': self.adjust_volumn_prompts}] input_prompt = { "role": "user", "content": human_input } self.adjust_volumn_message.append(input_prompt) response = self.client.chat.completions.create( model=self.model['adjust_volumn_model'], messages=self.adjust_volumn_message, stream=False, stop=["}"], timeout=10 ).json() # 解析并提取 question_function response = json.loads(response) adjust_volumn_result = response['choices'][0]['message']['content'] adjust_volumn_result = adjust_volumn_result + "}" adjust_volumn_result = re.findall(r'\{(.*?)\}', adjust_volumn_result)[0] print("adjust_volumn_result:",adjust_volumn_result) except Exception as e: self.logger.log_error("音量调节模型有问题") try: # response = requests.post('http://127.0.0.1:5000/adjust_volumn', data=adjust_volumn_result) response = requests.post( 'http://127.0.0.1:5000/adjust_volume', json={'adjust_volumn_result': adjust_volumn_result} ) # # 判断符号 # if "+" in adjust_volumn_result: # volume_change = int(re.search(r'\+(\d+)', adjust_volumn_result).group(1)) # print(f"增加音量: {volume_change}%") # os.system(f"/usr/bin/amixer -D pulse sset Master {volume_change}%+") # elif "-" in adjust_volumn_result: # volume_change = int(re.search(r'-(\d+)', adjust_volumn_result).group(1)) # print(f"减少音量: {volume_change}%") # os.system(f"/usr/bin/amixer -D pulse sset Master {volume_change}%-") # elif "=" in adjust_volumn_result: # volume_set = int(re.search(r'=(\d+)', adjust_volumn_result).group(1)) # print(f"设置音量为: {volume_set}%") # os.system(f"/usr/bin/amixer -D pulse sset Master {volume_set}%") # else: # print("无法识别的音量调节指令") except Exception as e: print(f"处理音量调节指令时发生错误: {e}") if operation == 'ylcx': volumn_result = self.get_volume() print("volumn_result:",volumn_result) try: # 使用正则表达式匹配音量百分比值 match = re.search(r'(\d+)%', volumn_result) if match: print("int(match.group(1)):",int(match.group(1))) match = int(match.group(1)) human_input = f"{human_input} {match}" print("human_input:",human_input) else: self.logger.log_error("无法提取出音量信息") # 未找到匹配 except Exception as e: print(f"提取音量时发生错误: {e}") if operation=='jttq': self.weather_info = self.get_info('/get_weather') print("weather_info",self.weather_info) time_result = datetime.now().strftime('%Y-%m-%d %H:%M:%S') time_result = time_result.split(' ')[1] human_input = f"天气提问：{human_input} 天气数据：{self.weather_info} (其中daily.0.fxDate为今日日期, 当前时间为{time_result}) 你需要结合geo_info和时间来分析当前用户提问的是接下来几天daily还是接下来几小时hourly还是当前天气now, 你需要综合分析并准确根据数据回答,不需要给分析过程，给出分析后的信息就可以,回复用户询问的相应时间点的天气就可以，回复需要简短，一句话说完" if operation=='dayh': massage_path = self.massage_status_languge['massage_path'] if massage_path in self.pathTypeMap: massage_path = self.pathTypeMap[massage_path] print(massage_path) # This will output: 循经直推法 else: massage_path=None human_input = f"{human_input} 目前按摩穴位:从{self.massage_status_languge['start_pos']}到{self.massage_status_languge['end_pos']} 目前按摩手法：{massage_path}" if operation=='tgdw': massage_path = self.massage_status_languge['massage_path'] if massage_path in self.pathTypeMap: massage_path = self.pathTypeMap[massage_path] print(massage_path) # This will output: 循经直推法 else: massage_path=None human_input = f"{human_input} 目前按摩穴位:从{self.massage_status_languge['start_pos']}到{self.massage_status_languge['end_pos']} 目前按摩手法：{massage_path} 需要跳过该区域" if operation == 'sftj': asyncio.run(self.send_message("massage_plan_start")) self.logger.log_info("进入手法推荐模块") def model_output(messages, model='qwen2.5-7b-instruct', stream=False, timeout=10): try: # 调用大模型接口 completion = self.client.chat.completions.create( model=model, messages=messages, stream=stream, timeout=timeout ) return completion except Exception as e: asyncio.run(self.send_message("JSON_error")) print(f"错误信息：{e}") return {"error": f"调用大模型时出错：{e}"} messages1=[ {'role': 'system', 'content': self.body_part_prompts}, {'role': 'user', 'content': f"提问：{human_input}"}, {'role': 'user', 'content': "请根据用户提问推断出用户想要按摩的部位及按摩头选择"}] # completion1=model_output(messages1) # print("completion1:",completion1) try: completion1 = model_output(messages1,model='qwen2.5-72b-instruct',stream=False) print(completion1.model_dump_json()) response_json = json.loads(completion1.model_dump_json()) content_str = response_json['choices'][0]['message']['content'] # print("content_str：",content_str) json_data = content_str.strip().strip('```json').strip() body_and_header = json.loads(json_data) self.logger.log_blue(f"body_and_header:{body_and_header}") except Exception as e: asyncio.run(self.send_message("JSON_error")) print("解析JSON格式部位失败") return "解析JSON格式部位失败" messages0=[ {'role': 'system', 'content': "你是中医智能理疗分析师，请从用户输入给出理由为什么需要推荐按摩手法"}, {'role': 'user', 'content': f"提问：{human_input}"}, {'role': 'user', 'content': f"中医推荐按摩部位：{body_and_header.get('body_part')}"}, {'role': 'user', 'content': "请根据用户提问给出需要推荐按摩手法的原因，原因分析请根据推荐部位进行"}] try: completion0=model_output(messages0,stream=True) full_content1 = '' for chunk in completion0: if chunk.choices: result = getattr(chunk.choices[0].delta, "content", None) if result: full_content1 += result instruction = {'message':result,'isReasoning':True} requests.post('http://127.0.0.1:5000/ai_respon', data=instruction) except Exception as e: print("生成原因失败") return "生成原因失败" massage_heads = { "指疗通络": ["背部", "肩颈", "腿部", "腰部"], "点阵按摩": ["背部", "肩颈", "腰部"], "滚滚刺疗": ["背部", "腿部"], "全能滚珠": ["背部"], "深部热疗": ["背部", "腹部", "腰部"], "温砭舒揉": ["背部", "腹部", "腰部"] } # 假设输入的部位和任务 body_part = body_and_header.get('body_part') task = body_and_header.get('choose_task') # 检查该部位是否能使用对应的按摩头 if body_part not in massage_heads.get(task, []): print(f"{task} 不可以在 {body_part} 上使用。") asyncio.run(self.send_message("header_not_use")) return f"{task} 不可以在 {body_part} 上使用。" messages2=[ {'role': 'system', 'content': self.message_retrieval_prompt}, {'role': 'user', 'content': f"提问：{human_input}"}, {'role': 'user', 'content': f"部位、按摩头：{body_and_header}"}, {'role': 'user', 'content': "请根据用户提问及提供的部位、按摩头，生成该部位的个性化中医理疗方案"}] try: completion2=model_output(messages2,stream=True) if isinstance(completion2, dict) and 'error' in completion2: print(f"模型调用失败：{completion2['error']}") return "无法获取理疗轨迹信息检索后消息" full_content = '' for chunk in completion2: if chunk.choices: result = getattr(chunk.choices[0].delta, "content", None) if result: full_content += result instruction = {'message':result,'isReasoning':False} requests.post('http://127.0.0.1:5000/ai_respon', data=instruction) asyncio.run(self.send_message("trajectory_generate")) # 输出结束后通知：轨迹生成开始 status_payload = { 'status': 'start', # 其他值可为 'fail' 或 'done' 'phase_message': '轨迹生成流程已启动' } requests.post('http://127.0.0.1:5000/trajectory_status', json=status_payload) except Exception as e: asyncio.run(self.send_message("JSON_error")) print("推荐方案失败") return "推荐方案失败" plan_body_part = body_and_header.get('body_part') print("plan_body_part:",plan_body_part) if plan_body_part == '肩颈': body_and_header['body_part'] = 'shoulder' body_part_prompts = self.shoulder_limit_prompts elif plan_body_part == '背部': body_and_header['body_part'] = 'back' body_part_prompts = self.back_limit_prompts elif plan_body_part == '腰部': body_and_header['body_part'] = 'waist' body_part_prompts = self.waist_limit_prompts elif plan_body_part == '腹部': body_and_header['body_part'] = 'belly' body_part_prompts = self.belly_limit_prompts elif plan_body_part == '腿部': body_and_header['body_part'] = 'leg' body_part_prompts = self.leg_limit_prompts else: print("获取到的部位有误") status_payload = { 'status': 'fail', 'phase_message': '轨迹生成流程已失败' } requests.post('http://127.0.0.1:5000/trajectory_status', json=status_payload) return "获取部位有误" messages3=[ {'role': 'system', 'content': body_part_prompts}, {'role': 'user', 'content': f"提问：{human_input}"}, {'role': 'user', 'content': f"部位、按摩头：{body_and_header}"}, {'role': 'user', 'content': "请根据用户提问及提供的部位、按摩头，生成该部位的中医理疗按摩手法轨迹"}] try: completion3 = model_output(messages3,model='qwen-plus', stream=False,timeout=60) print(completion3.model_dump_json()) response_json = json.loads(completion3.model_dump_json()) content_str = response_json['choices'][0]['message']['content'] # print("content_str：",content_str) json_data = content_str.strip().strip('```json').strip() massage_plan = json.loads(json_data) self.logger.log_blue(f"massage_plan:{massage_plan}") except Exception as e: status_payload = { 'status': 'fail', 'phase_message': '轨迹生成流程已失败' } requests.post('http://127.0.0.1:5000/trajectory_status', json=status_payload) asyncio.run(self.send_message("JSON_error")) print("解析JSON格式轨迹失败") return "解析JSON格式轨迹失败" try: plan_timestamp = datetime.now().strftime("%Y%m%d%H%M%S") plan_name = f"{self.task_mapping[massage_plan['choose_task']]}-{self.body_mapping[massage_plan['body_part']]}-{massage_plan['title']}{plan_timestamp}" plan_selection = f"按摩头以及部位选择:{self.task_mapping[massage_plan['choose_task']]}-{self.body_mapping[massage_plan['body_part']]}" massage_plan['can_delete'] = True self.vtxdb.set("massage_plan", plan_name, massage_plan) instruction = {'message':f"____\n{plan_selection}\n"} requests.post('http://127.0.0.1:5000/ai_respon', data=instruction) instruction = {'message':f"____\n按摩方案名称: {plan_name}"} requests.post('http://127.0.0.1:5000/ai_respon', data=instruction) asyncio.run(self.send_message("massage_plan_finish")) status_payload = { 'status': 'done', 'phase_message': '轨迹生成流程已完成' } requests.post('http://127.0.0.1:5000/trajectory_status', json=status_payload) except Exception as e: status_payload = { 'status': 'fail', 'phase_message': '轨迹生成流程已失败' } requests.post('http://127.0.0.1:5000/trajectory_status', json=status_payload) asyncio.run(self.send_message("JSON_error")) print(f"检查参数服务器key格式或者没写进参数服务器{e}") return "按摩手法记录失败，请重新生成" # 如果操作为 jtrq、xzsj 或 qthd，则生成模型响应 if operation in ['jtrq', 'xzsj', 'qthd','lddx', 'wddx', 'dldw', 'pldw', 'cjdx', 'amjd','amdt','ambw','gnjs','cqjm','ylcx','zsdw','jtxw','dayh','tgdw']: self.chat_message = [msg for msg in self.chat_message if msg.get('role') != 'user'] input_prompt = { "role": "user", "content": human_input } self.chat_message.append(input_prompt) # print("self.chat_message:",self.chat_message) if operation=='qthd' or operation == 'jtxw': if self.deep_thinking_flag == True and self.search_flag == False: response_generator = self.client_deep.chat.completions.create( model=self.model['deep_thinking_model'], messages=self.chat_message, stream=True, timeout=10 ) elif self.deep_thinking_flag == False and self.search_flag == True: user_id = os.uname()[1] response_generator = self.client_search.chat_completion(user_id=user_id, query=human_input,callback=self.finished_callback) elif self.deep_thinking_flag == True and self.search_flag == True: user_id = os.uname()[1] response_generator = self.client_search_deep.chat_completion(user_id=user_id, query=human_input,callback=self.finished_callback) else: response_generator = self.client.chat.completions.create( model=self.model['chat_model'], messages=self.chat_message, stream=True, timeout=10 ) else: response_generator = self.client.chat.completions.create( model=self.model['chat_model'], messages=self.chat_message, stream=True, timeout=10 ) if operation=='dayh': human_input=f"{human_input} 按摩过程调整：时间,加长" if operation=='tgdw': human_input=f"{human_input} 按摩过程调整：时间,跳过" return { 'chat_message': human_input, 'response_generator': response_generator } if operation == 'jttq': input_prompt = { "role": "user", "content": human_input } self.weather_message.append(input_prompt) if self.deep_thinking_flag == True: response_generator = self.client_deep.chat.completions.create( model=self.model['deep_thinking_model'], messages=self.weather_message, stream=True, timeout=10 ) else: response_generator = self.client.chat.completions.create( model=self.model['chat_model'], messages=self.weather_message, stream=True, timeout=10 ) self.weather_message.pop() return { 'chat_message': "询问天气", 'response_generator': response_generator } # 如果操作不在 jtrq、xzsj、qthd 之中，则返回其他操作的结果 return self.operations.get(operation, self.operations['qthd']) async def send_message(self,data): uri = "ws://localhost:8766" # 或替换成目标服务器地址 try: async with websockets.connect(uri) as websocket: await websocket.send(data) # 发送消息 print("消息已发送:", data) # 打印发送的消息 except websockets.exceptions.ConnectionClosedOK as e: # 连接正常关闭的异常处理 print(f"连接已关闭：{e}") except websockets.exceptions.WebSocketException as e: # 处理其他 WebSocket 异常 print(f"WebSocket 错误: {e}") except Exception as e: # 捕获其他异常 print(f"发生错误: {e}") def stop_thread(self): if hasattr(self, 'thread') and self.thread and self.thread.is_alive(): self.stop_event.set() self.thread.join(timeout=1) print("旧线程安全终止") def finished_callback(self,event_type,event_title): # if event_type=="node_finished" and event_title=="massage_method_suggestion_llm": # self.suggestion_flag=True # print("self.suggestion_flag:",self.suggestion_flag) # print("执行到播报结束...") # if event_type=="node_started" and event_title=="吹水": # self.suggestion_flag=False # print("self.suggestion_flag:",self.suggestion_flag) # print("执行到开始json...") if event_type=="node_started" and event_title=="深度思考轨迹": # self.suggestion_deekseek_flag=True # print("self.suggestion_deekseek_flag:",self.suggestion_deekseek_flag) print("执行到开始深度思考轨迹...") # if event_type=="node_finished" and event_title=="深度思考轨迹": # self.suggestion_deekseek_flag=False # print("self.suggestion_flag:",self.suggestion_deekseek_flag) # print("执行到结束深度思考轨迹...") def classify(self, human_input): """ 输入语句并与LLM交流，进行一次模型回复，返回结果并统计时间 """ start_time = time.time() # 开始计时 self.classify_message = [{'role': 'system', 'content': self.classify_prompts}] input_prompt = { "role": "user", "content": human_input } self.classify_message.append(input_prompt) try: # LLM交流并获取回复 response = self.client.chat.completions.create( model=self.model['classify_model1'], messages=self.classify_message, stream=False, stop=["}"], timeout=10 ).json() # 解析并提取 question_function response = json.loads(response) question_function = response['choices'][0]['message']['content'] question_function = question_function + "}" question_function = re.findall(r'\{(.*?)\}', question_function)[0] print(f"classifyProcessed question_function: {question_function}") except Exception as e: self.logger.log_error(f"Error processing question_function: {e}") return "cwsb" total_time = time.time() - start_time print(f"Total time for processing: {total_time:.2f} seconds") return question_function def correct_classify(self, human_input): # time.sleep(1) """ 输入语句并与LLM交流，进行一次模型回复，返回结果并统计时间 """ start_time = time.time() # 开始计时 self.classify_message2 = [{'role': 'system', 'content': self.classify_prompts2}] input_prompt = { "role": "user", "content": human_input } self.classify_message2.append(input_prompt) try: # LLM交流并获取回复 response = self.client.chat.completions.create( model=self.model['classify_model2'], messages=self.classify_message2, stream=False, stop=["}"], timeout=10 ).json() # 解析并提取 question_function response = json.loads(response) question_function = response['choices'][0]['message']['content'] question_function = question_function + "}" question_function = re.findall(r'\{(.*?)\}', question_function)[0] print(f"correct_classifyProcessed question_function: {question_function}") except Exception as e: self.logger.log_error(f"Error processing question_function: {e}") return "cwsb" total_time = time.time() - start_time print(f"Total time for processing: {total_time:.2f} seconds") return question_function def classify_or_correct(self, human_input): """ 生成器：先 yield 第一个返回值，再 yield 全部返回值。 """ result_queue = queue.Queue() final_results = {} lock = threading.Lock() first_result_sent = threading.Event() def wrapper(func, name): def inner(): try: result = func(human_input) except Exception as e: result = "cwsb" self.logger.log_error(f"[{name}] Error: {e}") with lock: final_results[name] = result if not first_result_sent.is_set(): result_queue.put((name, result)) first_result_sent.set() return inner # 启动两个线程 thread1 = threading.Thread(target=wrapper(self.classify, "classify")) thread2 = threading.Thread(target=wrapper(self.correct_classify, "correct_classify")) thread1.start() thread2.start() # yield 第一结果（谁先到谁 yield） first_name, first_result = result_queue.get(timeout=12) yield { "stage": "first", "function": first_name, "result": first_result } # 等两个都完成 thread1.join() thread2.join() # yield 所有结果 yield { "stage": "all", "results": final_results } def music_keyword_retrieve(self,human_input): try: self.retrieve_message = [{'role': 'system', 'content': self.retrieve_prompts}] input_prompt = { "role": "user", "content": human_input } self.retrieve_message.append(input_prompt) response = self.client.chat.completions.create( model=self.model['retrieve_model'], messages=self.retrieve_message, stream=False, stop=["}"], timeout=10 ).json() # 解析并提取 question_function response = json.loads(response) question_retrieve = response['choices'][0]['message']['content'] print(question_retrieve) return question_retrieve except Exception as e: self.logger.log_error("获得音乐关键词失败") def get_info(self,endpoint,params=None): base_url = 'http://127.0.0.1:5000' url = f"{base_url}{endpoint}" try: # 发送 GET 请求 response = requests.get(url, params=params) # 如果请求成功，返回 JSON 数据 if response.status_code == 200: return response.json() else: print(f"Error: {response.status_code}, {response.text}") return None except requests.RequestException as e: print(f"请求失败: {e}") return None def error_answer(self, **kwargs): human_input = kwargs.get('human_input') self.chat_message1 = [{'role': 'system', 'content': self.chat_prompts1}] input_prompt = { "role": "user", "content": human_input } self.chat_message1.append(input_prompt) response_generator = self.client.chat.completions.create( model=self.model['chat_model1'], messages=self.chat_message1, stream=True, timeout=10 ) return { 'response_generator': response_generator } def chat(self,human_input): self.logger.log_yellow("-----------------------------------------------------") try: return_dict = {} gen = self.classify_or_correct(human_input) # 第一阶段：抢先拿到第一个结果 first_finish = next(gen) self.logger.log_info(f"先完成返回:{first_finish}") question_function=first_finish['result'] first_source = first_finish['function'] # print("提取函数后值：",question_function) self.logger.log_blue(f"提取函数后值：{question_function}") valid_functions = [ 'ksam', 'tzam', 'ldzd', 'ldjx', 'jtrq', 'xzsj', 'jttq', 'jtxw', 'dlzd', 'dljx', 'pljk', 'pljd', 'cjlz', 'cjlj', 'zszj', 'zsjx', 'gbzx', 'tsgd','jdgd','dayh','tgdw', 'lddx', 'wddx', 'dldw', 'pldw', 'cjdx', 'amjd', 'amdt', 'ambw', 'gnjs', 'cqjm','ylcx','yltj', 'zsdw', 'wdzj', 'wdjx', 'qthd', 'cwsb', 'bfyy', 'tzbf', 'ssyy', 'xsyy', 'sftj'] if question_function not in valid_functions: question_function = 'cwsb' return_dict.update({'question_function': question_function}) # print("return_dict:",return_dict) time4=time.time() # self.logger.log_blue(f"一层大模型分类时间：{time4-time3}") kwargs = {'human_input': human_input} if question_function != 'cwsb': time5=time.time() chat_response = self.handle_request(question_function, **kwargs) print("chat_response:",chat_response) if question_function=='bfyy': question_retrieve=self.music_keyword_retrieve(human_input) chat_response=f"音乐播放调整：{question_retrieve}" time6=time.time() self.logger.log_blue(f"二层时间：{time6-time5}") # 处理返回的多种信息 if isinstance(chat_response, dict): chat_message = chat_response.get('chat_message', '') response_generator = chat_response.get('response_generator', '') return_dict.update({'chat_message': chat_message, 'response_generator': response_generator}) else: return_dict.update({'chat_message': chat_response}) else: return_dict.update({'chat_message': '我没有理解您的意思，请重新提问。'}) self.logger.log_yellow(f"改正前的return_dict:{return_dict}") yield return_dict if first_source == "classify": try: all_finish = next(gen) classify_result = all_finish['results'].get('classify') correct_result = all_finish['results'].get('correct_classify') self.logger.log_info(f"最后完成返回:{all_finish}") if classify_result != correct_result: self.logger.log_info("两个模型结果不一致，使用correct_classify更新返回") question_function=correct_result # 重新请求正确分类的处理 chat_response = self.error_answer(**kwargs) print("修正后chat_response:",chat_response) # 更新 return_dict if correct_result == 'bfyy': question_retrieve = self.music_keyword_retrieve(human_input) chat_response = f"音乐播放调整：{question_retrieve}" if isinstance(chat_response, dict): if question_function in ['ksam','tzam','ldzd','ldjx','wdzj','wdjx','dlzd','dljx','pljk','pljd','cjlz','cjlj','zszj','zsjx','gbzx','dayh','tgdw','jtrq','xzsj','jttq','jtxw','qthd','bfyy','tzbf','ssyy','xsyy']: return_dict.update({ 'question_function': question_function, 'chat_message': self.operations1.get(question_function), 'response_generator': chat_response.get('response_generator', ''), 'source': 'correct_classify' }) else: return_dict.update({ 'question_function': question_function, 'chat_message': kwargs.get('human_input'), 'response_generator': chat_response.get('response_generator', ''), 'source': 'correct_classify' }) else: return_dict.update({'chat_message': chat_response}) # 第二次 yield：更新后的 return_dict self.logger.log_yellow(f"改正后的return_dict:{return_dict}") yield return_dict except Exception as e: self.logger.log_error(f"二阶段处理出错: {e}") except Exception as e: self.logger.log_error(f"二层模型问题: {e}") if __name__ == '__main__': dashscopeclient = DashscopeClient() while True: # 从终端输入获取 human_input human_input = input("请输入问题：") if human_input.lower() in ['exit', 'quit', 'q']: print("程序结束") break time1 = time.time() return_gen = dashscopeclient.chat(human_input) # print(return_dict) return_dict=next(return_gen) print("return_dict:",return_dict) def handle_all_finish(): try: update_dict = next(return_gen) print("update_dict:",update_dict) if 'source' in update_dict and update_dict['source'] is not None: flag1=True except StopIteration: pass # 生成器已结束 except Exception as e: print("处理最终结果时出错:",e) threading.Thread(target=handle_all_finish).start() ######流式大模型输出的时输出方式######## if isinstance(return_dict, str): pass else: response_generator = return_dict.get('response_generator', '') punctuation_marks_regex = r'[。,，；！？]' full_content = '' last_index = 0 full_reasoning_content = '' for response in response_generator: # if dashscopeclient.suggestion_flag==False: # print(response) if response.choices: reasoning_result = getattr(response.choices[0].delta, "reasoning_content", None) # print("reasoning_content:",reasoning_result) result = getattr(response.choices[0].delta, "content", None) # print("content:",reasoning_result) if reasoning_result: full_reasoning_content += reasoning_result print(reasoning_result) if result: full_content += result punctuation_indices = [m.start() for m in re.finditer(punctuation_marks_regex, full_content)] for index in punctuation_indices: if index > last_index: accumulated_text = full_content[last_index:index+1] last_index = index + 1 if accumulated_text.strip(): print(accumulated_text.strip()) # else: # print("结束了") # break ################################### time2 = time.time() print(time2-time1)