LCA_LLM_application/Retrieval_new/embedding.py

import os
import json
from flask import Flask, request, make_response
from logzero import logger
from functools import lru_cache

from utils import *
from local_encoder import EmbeddingModel

# current_path = os.path.dirname(os.path.abspath(__file__)) # for local
current_path = os.getcwd()  # for docker
logger.info(f"{current_path}")

app = Flask(__name__)
os.environ["CUDA_DEVICE_ORDER"] = 'PCI_BUS_ID'
os.environ['CUDA_VISIBLE_DEVICES'] = '0,1'


path = "/home/zhangxj/models/bge-large-zh-v1.5"
model = EmbeddingModel(path)

@lru_cache()
def process_and_embed(sentence):  
    # 字符串全部转化为中文处理
    sentence = translate(sentence)

    # 中文预处理
    clean_text = preprocess_zh(sentence)
    processed_text = get_noun_zh(clean_text)

    # 如果处理后为空，使用原始文本
    if not processed_text.strip():
        processed_text = sentence
        
    # 获取向量编码
    embeddings = model.get_embeddings(processed_text)
    return embeddings

@app.route('/embedding/', methods=["POST"])
def run_cls():
    resp_info = dict()
    if request.method == "POST":
        sentences = request.json.get('sentences')
        if sentences is not None and len(sentences) != 0:
            logger.info(sentences)
            resp_info["code"] = 200
            resp_info["data"] = process_and_embed(sentences)
        else:
            resp_info["msg"] = "Input is None, please check !"
            resp_info["code"] = 406
    resp = make_response(json.dumps(resp_info))
    resp.status_code = 200
    return resp

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=5163, debug=False)
    # res = process_and_embed("土豆")
    # print(res)