QwenModel/text_analysis/bak/views.py0802_3

# coding:utf8
import os, sys
import io

sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding='utf8')
cur_dir = os.path.dirname(os.path.abspath(__file__)) or os.getcwd()
par_dir = os.path.abspath(os.path.join(cur_dir, os.path.pardir))
sys.path.append(cur_dir)
sys.path.append(par_dir)
import json
from django.http import HttpResponse
from text_analysis.tools import to_kafka
from text_analysis.cusException import promptLen_Exception
from django.views.decorators.csrf import csrf_exempt
from log_util.set_logger import set_logger
from openai import OpenAI
logging = set_logger('logs/results.log')
import traceback
import queue
import requests
from text_analysis.tools.tool import get_content,parse_gptResult
import uuid
import time
from kazoo.client import KazooClient
from kazoo.protocol.states import EventType
# openai_api_key = "EMPTY"
# openai_api_base = "http://10.0.32.225:9000/v1"
# client = OpenAI(api_key=openai_api_key,base_url=openai_api_base)
task_queue = queue.Queue()
stop_dict={}

@csrf_exempt
def QwenModel(request):
    if request.method == 'POST':
        try:
            # txt=request.body.encode("utf-8")
            raw_data = json.loads(request.body)
            task_queue.put(raw_data)
            return HttpResponse(json.dumps({"code": 1, "msg": "请求正常！"}, ensure_ascii=False))
        except:
            logging.error(traceback.format_exc())
            return HttpResponse(json.dumps({"code": 0, "msg": "请求json格式不正确！"}, ensure_ascii=False))
    else:
        return HttpResponse(json.dumps({"code": 0, "msg": "请求方式错误，改为post请求"}, ensure_ascii=False))


def Qwen():
    while True:
        if task_queue.qsize() > 0:
            try:
                logging.info("取任务队列长度{}".format(task_queue.qsize()))
                raw_data = task_queue.get()
                output = raw_data["output"]
                res_tmp = {key: "" for key in output}
                if "id" in res_tmp.keys():
                    res_tmp["id"] = str(uuid.uuid4())
                task_id=raw_data["scenes_id"]
                task_version=raw_data["version"]
                if "data" not in raw_data.keys():
                    logging.info("任务缺少data—{}".format(raw_data))
                    raw_data["result"] = {"successCode": "0", "errorLog": "", "results": "", "status": 2,"message": "未配置data内容"}
                    res_tmp_json = json.dumps(res_tmp, ensure_ascii=False)
                    raw_data["result"]["results"] = res_tmp_json
                    to_kafka.send_kafka(raw_data, logging)
                else:
                    logging.info("任务数据为：{}".format(raw_data))
                    logging.info("当前version信息为：{}".format(stop_dict))
                    if task_id in stop_dict.keys() and task_version!=stop_dict[task_id]["version"]:
                        logging.info("已暂停任务，过滤掉。{}".format(raw_data))
                        continue
                    data = get_content(raw_data, logging)
                    if len(data["prompt"])>=10000:
                        raise promptLen_Exception
                    # logging.info("请求信息为{},解析后模型请求为{}".format(raw_data,data))
                    url="http://10.0.32.225:9000/v1/chat/completions"
                    headers = {
                        "Content-Type": "application/json;charset=UTF-8"
                    }
                    payload = json.dumps({
                        "model":"Qwen2-72B-Instruct-GPTQ-Int4",
                        "messages": [{"role": "user", "content": data["prompt"]}],
                        "temperature": float(data["temperature"]),
                        "top_p": float(data["top_p"]),
                        "n": int(data["n"])
                    })
                    # logging.info("prompt为{}".format(data["prompt"]))
                    response = requests.request("POST", url, headers=headers, data=payload,timeout=180)
                    logging.info("Prompt为：{}—Qwen返回值：{}-{}".format(data["prompt"],response,response.text))
                    d = json.loads(response.text)
                    if response.status_code==400 and "maximum context length" in d["message"]:
                        raise promptLen_Exception
                    result = d['choices'][0]['message']['content']
                    if result:
                        #添加 0是文本，1是json格式
                        fieldType = raw_data["input"]['fieldType']
                        if fieldType == 0:
                            res_tmp["content"] = result
                            res_tmp_json = json.dumps(res_tmp, ensure_ascii=False)
                            raw_data["result"] = {"successCode": "1", "errorLog": "", "results": res_tmp_json,"status":1,"message":"成功"}
                        else:
                            res=parse_gptResult(res_tmp,result)
                            if res:
                                res_tmp_json = json.dumps(res, ensure_ascii=False)
                                raw_data["result"] = {"successCode": "1", "errorLog": "", "results": res_tmp_json,"status":1,"message":"成功"}
                            else:
                                res_tmp_json = json.dumps(res_tmp, ensure_ascii=False)
                                raw_data["result"] = {"successCode": "0", "errorLog": "GPT返回结果不符合预期", "results": res_tmp_json,"status":2,"message":"GPT返回结果不符合预期"}
                    else:
                        logging.info("模型返回值正常但为空-{}".format(raw_data))
                        res_tmp_json = json.dumps(res_tmp, ensure_ascii=False)
                        raw_data["result"] = {"successCode": "1", "errorLog": "", "results": res_tmp_json, "status": 1,"message": "成功"}

                    logging.info(raw_data)
                    to_kafka.send_kafka(raw_data, logging)
            except promptLen_Exception:
                logging.info("文本长度超过模型限制-{}".format(raw_data))
                res_tmp_json = json.dumps(res_tmp, ensure_ascii=False)
                raw_data["result"] = {"successCode": "0", "errorLog": "", "results": res_tmp_json, "status": 2,"message": "文本长度超过模型限制"}
                to_kafka.send_kafka(raw_data, logging)

            except:
                res_tmp_json = json.dumps(res_tmp, ensure_ascii=False)
                raw_data["result"] = {"successCode": "0", "errorLog": traceback.format_exc(), "results": res_tmp_json,"status":2,"message":"异常"}
                logging.info("调用Qwen失败{}-{}".format(raw_data, traceback.format_exc()))
                to_kafka.send_kafka(raw_data, logging)

        else:
            logging.info("暂无任务，进入休眠--")
            time.sleep(10)

def zk_monitoring():
    try:
        #线上环境
        zk = KazooClient(hosts='172.18.1.146:2181,172.18.1.147:2181,172.18.1.148:2181')
        #测试环境
        # zk = KazooClient(hosts='172.16.12.55:2181,172.16.12.56:2181,172.16.12.57:2181')
        zk.start()
        # 设置监听器
        @zk.DataWatch("/analyze")
        def watch_node(data, stat, event):
            if event is not None and event.type == EventType.CHANGED:
                data, stat = zk.get("/analyze")
                logging.info("执行删除操作：{}".format(data))
                d = json.loads(data)
                id = d["scenes_id"]
                stop_dict[id] = {}
                stop_dict[id]["version"] = d["version"]
                stop_dict[id]["operation"] = d["operation"]
        # 保持程序运行以监听节点变化
        try:
            while True:
                time.sleep(1)
        except:
            logging.info("Stopping...")
        # 关闭连接
        zk.stop()
        zk.close()
    except:
        logging.error(traceback.format_exc())