python_ai/pytorch纠错模型服务化部署macbert4csc.py

import threading
# from checkPlaceName import checkPlaceName
# from checkRepeatText import checkRepeatText
# from checkCompanyName import checkCompanyName
# from checkDocumentError import checkDocumentError
# from checkTitleName import checkTitleName
# from myLogger import outLog
# import time
# def run_check_company_name(filename,user_id):
#     for i in checkCompanyName(filename,user_id):
#         pass

# def run_get_document_error(filename,user_id):
#     for i in checkDocumentError(filename,user_id):
#         pass
# def runcheckTitleName(filename,user_id):
#     for i in checkTitleName(filename,user_id):
#         pass
# def runcheckRepeatText(filename,user_id):
#     for i in checkRepeatText(filename,user_id):
#         pass
# def runcheckPlaceName(filename,user_id):
#     for i in checkPlaceName(filename,user_id):
#         pass

# def get(user_id):
#     time.sleep(5)
#     while True:
#         if outLog.is_done(user_id):
#             break
#         q = outLog.get_queueData(user_id)
#         if q:
#             text = q.pop(0)
#             print(text)
#     print("打印结束")

# filename = "17.docx"

# # 创建线程getapp
# thread1 = threading.Thread(target=run_check_company_name, args=(filename,"1"))
# thread2 = threading.Thread(target=run_get_document_error, args=(filename,"1"))
# thread3 = threading.Thread(target=runcheckTitleName, args=(filename,"1"))
# thread4 = threading.Thread(target=runcheckRepeatText, args=(filename,"1"))
# thread5 = threading.Thread(target=runcheckPlaceName, args=(filename,"1"))
# thread6 = threading.Thread(target=get, args=("1",))
# thread1 = threading.Thread(target=getapp, args=(filename,))
# thread2 = threading.Thread(target=getapp, args=(filename,))
# thread3 = threading.Thread(target=getapp, args=(filename,))
# thread4 = threading.Thread(target=getapp, args=(filename,))
# thread5 = threading.Thread(target=getapp, args=(filename,))
# thread6 = threading.Thread(target=getapp, args=("1",))
# # 启动线程
# thread1.start()
# thread2.start()
# thread3.start()
# thread4.start()
# thread5.start()
# thread6.start()
# # 等待线程完成
# thread1.join()
# thread2.join()
# thread3.join()
# thread4.join()
# thread5.join()
# thread6.join()
# print("Both tasks completed.")
# from pycorrector.macbert.macbert_corrector import MacBertCorrector

# m = MacBertCorrector("models")
# for i in range(10):
#     i = m.correct("行政捡查是行政机关覆行政府职能、管理经济社会事务的重要方式，开展计划统筹是行政检查控总量、提质效的重要措施和手段，直接影响改革或得感和社会满意度")
#     print(i)
# import re
# import json
# import json_repair
# import math
# import os
# import platform
# import torch
# import torch_npu
# import operator
# from torch_npu.contrib import transfer_to_npu
# torch_device = "npu:4" # 0~7
# torch.npu.set_device(torch.device(torch_device))
# torch.npu.set_compile_mode(jit_compile=False)
# from transformers import BertTokenizerFast,BertForMaskedLM
# # option = {}
# # option["NPU_FUZZY_COMPILE_BLACKLIST"] = "Tril"
# # torch.npu.set_option(option)
# print("torch && torch_npu import successfully")

# DEFAULT_CKPT_PATH = 'macbert4csc'
# #models=----macbert4csc-base-chinese
# model = BertForMaskedLM.from_pretrained(
#     DEFAULT_CKPT_PATH,
#     torch_dtype=torch.float16,
#     device_map=torch_device
# ).npu().eval()
# tokenizer = BertTokenizerFast.from_pretrained(DEFAULT_CKPT_PATH)
# def get_errors(corrected_text, origin_text):
#     sub_details = []
#     for i, ori_char in enumerate(origin_text):
#         if ori_char in [' ', '“', '”', '‘', '’', '琊', '\n', '…', '—', '擤']:
#             # add unk word
#             corrected_text = corrected_text[:i] + ori_char + corrected_text[i:]
#             continue
#         if i >= len(corrected_text):
#             continue
#         if ori_char != corrected_text[i]:
#             if ori_char.lower() == corrected_text[i]:
#                 # pass english upper char
#                 corrected_text = corrected_text[:i] + ori_char + corrected_text[i + 1:]
#                 continue
#             sub_details.append((ori_char, corrected_text[i], i, i + 1))
#     sub_details = sorted(sub_details, key=operator.itemgetter(2))
#     return corrected_text, sub_details
#     result = []
# def getapp(gettext):
#     result = []
#     batchNum = 20
#     sentences = re.split(r'[。\n]', gettext)
#     # 去掉空字符
#     sentences = [sentence.strip() for sentence in sentences if sentence.strip()]
#     # 计算总字符数
#     total_chars = len(sentences)

#     # 计算有多少份
#     num_chunks = math.ceil(total_chars / batchNum)

#     # 按batchNum字为一份进行处理
#     chunks = [sentences[i:i + batchNum] for i in range(0, total_chars, batchNum)]
#     # 打印每一份的内容
#     err = []
#     for i, chunk in enumerate(chunks):
#         inputs = tokenizer(chunk, padding=True, return_tensors='pt').to(torch_device)
#         with torch.no_grad():
#             outputs = model(**inputs)
#         for id, (logit_tensor, sentence) in enumerate(zip(outputs.logits, chunk)):
#             decode_tokens_new = tokenizer.decode(
#                 torch.argmax(logit_tensor, dim=-1), skip_special_tokens=True).split(' ')
#             decode_tokens_new = decode_tokens_new[:len(sentence)]
#             if len(decode_tokens_new) == len(sentence):
#                 probs = torch.max(torch.softmax(logit_tensor, dim=-1), dim=-1)[0].cpu().numpy()
#                 decode_str = ''
#                 for i in range(len(sentence)):
#                     if probs[i + 1] >= 0.7:
#                         decode_str += decode_tokens_new[i]
#                     else:
#                         decode_str += sentence[i]
#                 corrected_text = decode_str
#             else:
#                 corrected_text = sentence
#             print(corrected_text)
        #outputs = model(**tokenizer(chunk, padding=True, return_tensors='pt').to(torch_device))
        # for ids, text in zip(outputs.logits, chunk):
        #     _text = tokenizer.decode(torch.argmax(ids, dim=-1), skip_special_tokens=True).replace(' ', '')
        #     corrected_text = _text[:len(text)]
        #     corrected_text, details = get_errors(corrected_text, text)
        #     print(text, ' => ', corrected_text, details)
        #     result.append((corrected_text, details))
        # for i, sent in enumerate(chunk):
        #     decode_tokens = tokenizer.decode(outputs[i], skip_special_tokens=True).replace(' ', '')
        #     corrected_sent = decode_tokens[:len(sent)]
        #     print(corrected_sent)
                # corrected_sents.append(corrected_sent)
# from flask import Flask, request, jsonify
# import threading
# import time
# import re
# import math
# from macbert_corrector import MacBertCorrector
# m = MacBertCorrector("macbert4csc")
# app = Flask(__name__)

# # 创建一个锁对象
# lock = threading.Lock()
# #多线程但是每次只处理一个请求，多余的请求需要排队
# @app.route('/taskflow/checkDocumentError', methods=['POST'])
# def process_request():
#     with lock:

#         data = request.get_json()
#         # print("data",data)
#         # 提取文本数据
#         text_data = data.get('data', {}).get('text', [])
#         # print(text_data)
#         # 处理文本数据，例如检查错误等
#         # 这里您可以添加实际的逻辑来检查文档错误
#         res = m.correct_batch(text_data)
#         # 示例：简单打印接收到的文本
#         # # 返回响应
#         return jsonify({"status": "success", "data": res}), 200

# if __name__ == '__main__':
#     app.run(threaded=True,port=5001)
from fastapi import FastAPI, Request
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
import uvicorn
from fastapi.responses import JSONResponse
from pydantic import BaseModel
app = FastAPI()
from macbert_corrector import MacBertCorrector
m = MacBertCorrector("macbert4csc")

class RequestData(BaseModel):
    data: dict
@app.post("/taskflow/checkDocumentError")
async def process_request(request: RequestData):
    global m
    # 提取文本数据
    text_data = request.data.get('text')
    # 处理文本数据，例如检查错误等
    # 这里您可以添加实际的逻辑来检查文档错误

    # print(text_data)
    # 处理文本数据，例如检查错误等
    # 这里您可以添加实际的逻辑来检查文档错误
    res = m.correct_batch(text_data)
    
    # 返回响应
    return JSONResponse(content={"status": "success", "data": res}, status_code=200)
if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=5001)