当前位置：首页 > article >正文

python 大模型验证码识别

article 2024/9/20 14:41:59

一、注册账号https://huggingface.co/

方便下载大模型，并配置Token

进入国内镜像站：https://hf-mirror.com/

我们现在以验证码识别为例：webapi.py

安装依赖：

https://webapi.py/

1. 安装依赖
pip install -U huggingface_hub
pip install transformers datasets torch

#pip uninstall torch 
#conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch

2. 设置环境变量
Linux
export HF_ENDPOINT=https://hf-mirror.com

Windows Powershell
$env:HF_ENDPOINT = "https://hf-mirror.com"

3.1 下载模型
huggingface-cli download --token 自已上面生成的token --resume-download zenoda/trocr-captcha-killer --local-dir trocr-captcha-killer

打开：https://hf-mirror.com/zenoda/trocr-captcha-killer/tree/main

打开：https://hf-mirror.com/zenoda/trocr-captcha-killer 复制运行代码

注意要将：cuda改成cpu

运行完整代码

from flask import Flask,request
from werkzeug.utils import secure_filename
import uuid
from transformers import VisionEncoderDecoderModel, TrOCRProcessor
from PIL import Image
import requests

app = Flask(__name__)

@app.route('/')
def hello_world():
    return 'Hello, World!'

@app.post('/upload')
def file_upload():
    f=request.files['file']
    uuid1 =str(uuid.uuid1())
    path=uuid1 + f.filename
    f.save(f"{secure_filename(f.filename)}")

    processor = TrOCRProcessor.from_pretrained("D:/工作空间/练习空间/PythonProjects/pythonProject1/trocr-captcha-killer")
    model = VisionEncoderDecoderModel.from_pretrained("D:/工作空间/练习空间/PythonProjects/pythonProject1/trocr-captcha-killer")
    #model.to('cuda')
    model.to('cpu')
    url = 'https://hf-mirror.com/datasets/zenoda/trocr-captcha-killer/resolve/main/106-1688354008849.png'
    image = Image.open(requests.get(url, stream=True).raw).convert("RGB")
    #generated_ids = model.generate(processor(image, return_tensors="pt").pixel_values.to('cuda'))
    generated_ids = model.generate(processor(image, return_tensors="pt").pixel_values.to('cpu'))
    predictText = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    print(predictText)
    return '上传成功!'

if __name__ == '__main__':
    app.run()