padlepadle 文档抽取问答系统构建镜像

骑着蜗牛向前跑

已于 2023-06-29 22:36:48 修改

阅读量121

点赞数

分类专栏：人工智能文章标签： python 开发语言 paddlepaddle 文档智能问答系统

于 2022-11-09 21:11:47 首次发布

本文链接：https://blog.csdn.net/yy_diego/article/details/127778062

版权

人工智能专栏收录该内容

10 篇文章 0 订阅

订阅专栏

文档抽取问答系统

https://github.com/PaddlePaddle/PaddleNLP/tree/develop/pipelines/examples/document-intelligence

镜像构建

FROM nvidia/cuda:11.6.0-cudnn8-devel-ubuntu20.04

RUN echo "==> Install common tools"  && apt update && \
    DEBIAN_FRONTEND=noninteractive apt install -y vim git ssh wget curl libgomp1 gcc g++ \
        tar mc libgomp1 libbz2-dev libssl-dev  \
        libncursesw5-dev libdb-dev libexpat1-dev zlib1g-dev liblzma-dev libgdbm-dev \
        libffi-dev openssl libreadline6-dev libsqlite3-dev liblzma-dev \
        sqlite3 ffmpeg libsm6 libxext6 && \
    wget https://www.python.org/ftp/python/3.7.5/Python-3.7.5.tgz  && \
    tar -zxvf Python-3.7.5.tgz  && \
    cd Python-3.7.5  && \
    ./configure  --prefix=/usr/local  && \
    make && make install && \
    cd / && rm -rf compile    

WORKDIR /root

COPY entrypoint.sh .
#COPY paddleocr .paddleocr
#COPY paddlenlp .paddlenlp

RUN echo "==> Install PaddlePaddle" && \
    git clone https://github.com/PaddlePaddle/PaddleNLP.git && \
    pip3 config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple && \
    pip3 install --upgrade pip && \
    pip3 install lanms==1.0.2 && \
    paddlenlp==2.4.1 && \
    python3 -m pip install paddlepaddle-gpu==2.3.2.post111 -f \
    https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html && \
    pip3 install -r /root/PaddleNLP/requirements.txt && \
    pip3 install paddle-pipelines==0.3  && \
    cd /usr/local/bin/ && ln -s python3 python 
    
CMD ["/bin/bash","entrypoint.sh"]

entrypoint.sh

#!/bin/bash

echo "Run document-intelligence service" 
cd /root/PaddleNLP/pipelines
export PIPELINE_YAML_PATH=rest_api/pipeline/docprompt.yaml
export QUERY_PIPELINE_NAME=query_documents
bash examples/document-intelligence/run_docprompt_server.sh

构建镜像

docker build -t padlepadle:02 .

启动镜像

docker run -p 8890:8891 -d --runtime=nvidia padlepadle:02 .

测试

curl --request POST --url 'http://127.0.0.1:8890/query_documents' -H "Content-Type: application/json"  --data '{"meta": {"doc": "https://bj.bcebos.com/paddlenlp/taskflow/document_intelligence/images/invoice.jpg", "prompt": ["发票号码是多少?", "校验码是多少?"]}}'

返回结果

{"meta":{"doc":"https://bj.bcebos.com/paddlenlp/taskflow/document_intelligence/images/invoice.jpg","prompt":["发票号码是多少?","校验码是多少?"]},"results":[[{"prompt":"发票号码是多少?","result":[{"value":"No44527206","prob":0.74,"start":2,"end":2}]}