kuddah / dr-doc-search

Converse with book - Built with GPT-3

Doc Search

Converse with a book (PDF)

See tweet for full demo.

---

Documentation: https://namuan.github.io/dr-doc-search

Source Code: https://github.com/namuan/dr-doc-search

PyPI: https://pypi.org/project/dr-doc-search/

---

Pre-requisites

• Tessaract OCR
• ImageMagick

Note: If you are using Windows, then make sure that you set the location of ImageMagick executable in the IMCONV environment variable.

# For example, if you have installed ImageMagick in PROGRAMFILES\ImageMagick-7.1.0-Q16-HDRI
set IMCONV="%PROGRAMFILES%\ImageMagick-7.1.0-Q16-HDRI\magick"

Installation

pip install dr-doc-search

Example Usage

There are two steps to use this application:

1. First, you need to create the index and generate embeddings for the PDF file. Here I'm using a PDF file generated from this page Parable of a Monetary Economy

Before running this, you need to set up your OpenAI API key. You can get it from OpenAI.

From version 1.5.0, you can skip OpenAI and use HuggingFace models to generate embeddings and answers.

export OPENAI_API_KEY=<your-openai-api-key>

The run the following command to start the training process:

dr-doc-search --train -i ~/Downloads/parable-of-a-monetary-economy-heteconomist.pdf

Use huggingface for generating embeddings:

dr-doc-search --train -i ~/Downloads/parable-of-a-monetary-economy-heteconomist.pdf --embedding huggingface

The training process generates some temporary files in the OutputDir/dr-doc-search/<pdf-name> folder under your home directory. Here is what it looks like:

 ~/OutputDir/dr-doc-search/parable-of-a-monetary-economy-heteconomist
$ tree
.
├── images
│ ├── output-1.png
│ ├── output-10.png
│ ├── output-11.png
...
│ └── output-9.png
├── index
│ ├── docsearch.index
│ └── index.pkl
├── parable-of-a-monetary-economy-heteconomist.pdf
└── scanned
    ├── output-1.txt
    ...
    └── output-9.txt

Note: It is possible to change the base of the output directory by providing the --app-dir argument.

2. Now that we have the index, we can use it to start asking questions.

dr-doc-search -i ~/Downloads/parable-of-a-monetary-economy-heteconomist.pdf --input-question "How did the attempt to reduce the debut resulted in decrease in employment?"

Or You can open up a web interface (on port :5006) to ask questions:

dr-doc-search --web-app -i ~/Downloads/parable-of-a-monetary-economy-heteconomist.pdf

To use huggingface model, provide the --llm argument:

dr-doc-search --web-app -i ~/Downloads/parable-of-a-monetary-economy-heteconomist.pdf --llm huggingface

There are more options for choose the start and end pages for the PDF file. See the help for more details:

dr-doc-search --help

Acknowledgements

• anton/@abacaj for the idea
• LangChain
• HoloViz Panel
• OpenAI

Development

• Clone this repository

• Requirements:

  • Python 3.7+
  • Poetry

• Create a virtual environment and install the dependencies

poetry install

• Activate the virtual environment

poetry shell

Validating build

make build

Release process

A release is automatically published when a new version is bumped using make bump. See .github/workflows/build.yml for more details. Once the release is published, .github/workflows/publish.yml will automatically publish it to PyPI.

Disclaimer

This project is not affiliated with OpenAI. The OpenAI API and GPT-3 language model are not free after the trial period.