Skip to content

Latest commit

 

History

History
129 lines (101 loc) · 5.76 KB

README.md

File metadata and controls

129 lines (101 loc) · 5.76 KB
Raw

Text Summarization Power Skill

This Power Skill uses the pre-trained HuggingFace/Facebook BART model deep learning model to generate a summary for an input text - English only.

This skill is ideal for:

  1. Exploring your data to summarize and categorise your documents in the data exploration phase
  2. Using in conjunction with Azure Text Analytics or a Named Entity Recognition (NER) model and Latent Dirichlet Allocation (LDA) topic modelling to further identify areas of interest

This example uses the Meantime News dataset, and some sample texts have been placed in the data folder.

This Power Skill includes a Visual Studio Code dev container. If you use the dev container then your development environment will be set up with all the required dependencies:

  • python 3
  • jupyter notebooks
  • tensorflow
  • Terraform
  • Docker client
  • Azure CLI

Requirements

In addition to the common requirements described in the root README.md file, this Power Skill requires access to a Text Analytics resource. This process will use NER to illustrate the entities extracted and its relevance to the summarized text.

This PowerSkill will use:

  • Docker
  • Azure Container Registry
  • Azure Blob storage container
  • Azure Web App Service - Linux Containers
  • Azure AI Search instance
  • Azure AI Services

High level Process

Alt

Quick deployment

In order to deploy everything, you can simply use bash and type

make deploy

and type 'yes' when prompted. You can test your deployment by using this Notebook

Run locally

This section describes how to get the sample working in stages and how it can be amended for your data.

  1. Data

    The first step is to view the sample data files here.

  2. Run the API locally

    The next step is to run the API locally and test the model against a test record. Make sure you rename the file .env.example file to .env and populate it with the relevant values, particularly a unique key. Again use a bash terminal.

    make local-API

    NB: the HuggingFace/Facebook BART model is 1.6Gb, this will be downloaded and then cached. It may take some time to start the application.

    Upon starting you will see the download of the BART model initiate. See below for example:

    DEBUG:urllib3.connectionpool:https://cdn-lfs.huggingface.co:443 "GET /facebook/bart-large-cnn/2ac2745c02ac987d82c78a14b426de58d5e4178ae8039ba1c6881eccff3e82f1 HTTP/1.1" 200 1625270765
Downloading:   1%

    You should see the following when it is ready:

    INFO:uvicorn.error:Uvicorn running on http://0.0.0.0:5000 (Press CTRL+C to quit)

    You can test your local API two different ways:

  3. Build the docker image

    Now we can containerise the application by building and running a docker image and test our inference API locally.

    NB: Ensure you stop the python application from the previous step otherwise you will have port clashes. Again in bash, run

    make run-skill

    Note, the HuggingFace/Facebook BART model is 1.6Gb. If you have ran the previous step, hopefully it will not have to download the model again. It can still take a minute to load the 1.6Gb into memory though.

    Again, you can test your local API hosted in docker two different ways:

Run in Azure

In order to deploy this into Azure, we need some to build some shared infrastructure - you may already have this from another PowerSkill. If you do, you can simply rename base.env.example to be base.env and fill in the values.

  1. Deploy shared infrastructure

    If not, you can deploy the foundation simple by running the following in bash

    make deploy-base

    At this point, Terraform has now created the following infrastructure and created a base.env file with the values.

    • Azure Container Registry to store our docker image
    • Azure AI Services to enrich the data
    • Azure AI Search
    • Azure Blob Storage to hold our data to seach

  2. Push the container to Azure Container Registry

    We can now push our container that we built earlier to the cloud. Type the following command in your bash terminal.

    make push-skill

  3. Deploy the container to an Azure Web App.

    We will deploy this as to an Azure App Service Web App running a container. Run the following command:

    make deploy-skill

    You will be prompted with:

    Do you want to perform these actions?
  Terraform will perform the actions described above.
  Only 'yes' will be accepted to approve.

    Type

    yes

  4. Test the index

    Now we are in a position to search on our most similar data, navigate to the Let's go and test the ACS index to summarise our text and also run NER.