modescope环境运行gte-qwen2-1.5b文本转向量模型

发表于 2025/02/10 20:06:16 [ai] 浏览次数：145

代码示例：

#pip install gevent
#pip install flask

# 导入Flask类库
import os.path
import threading
import uuid

import requests
from flask import Flask, request

from gevent import pywsgi

import os
os.environ['MODELSCOPE_CACHE'] = './model_data'

# 向量模型
from sentence_transformers import SentenceTransformer
from modelscope import snapshot_download
# 自动下载模型
model_dir = snapshot_download("iic/gte_Qwen2-1.5B-instruct")

model = SentenceTransformer(model_dir, trust_remote_code=True)
# In case you want to reduce the maximum length:
# model.max_seq_length = 8192
model.max_seq_length = 2048



# 创建应用实例
app = Flask(__name__)


# 视图函数（路由）
@app.route("/")
def hello_world():
    return "Hello, World!"


@app.route("/text2vec", methods=["POST"])
def text2vec():
    # 在调用thread之前打印当前线程信息
    print("当前线程", threading.enumerate())

    text = request.form.get("text")
    query = request.form.get("query")
    result ={"status": 0}
    if query=='1':
         print("query vec")
         result["vec"] = model.encode([text], prompt_name="query")[0].tolist()
    else: 
         result["vec"] = model.encode([text])[0].tolist()
    
    return result



@app.route("/texts2vecs", methods=["POST"])
def texts2vecs():
    # 在调用thread之前打印当前线程信息
    print("当前线程", threading.enumerate())
    # 从请求中获取 JSON 数据
    request_data = request.json
    
    # 假设请求数据是一个包含数字的列表
    data_list = request_data.get('texts', [])
    result ={"status": 0}

    result["vec"] = model.encode(data_list).tolist()
    
    return result

# 启动服务
if __name__ == '__main__':
    # threaded=False 使用单线程模式，处理完一个请求再处理下一个请求
    # app.run(host='0.0.0.0', port=9012, threaded=False, debug=False)

    app.config['JSON_AS_ASCII'] = False
    app.config['threaded'] = False
    app.config['debug'] = False
    port = 9522
    # app.run(host='0.0.0.0', port=9012, threaded=True, debug=False)
    print("端口:", port)
    server = pywsgi.WSGIServer(('0.0.0.0', port), app)
    server.serve_forever()