Sample of using OpenAI GPT API with embeddings

Overview

• This repository is a sample code based on openapi's examples pages
• The content is the same, but we have made it possible to search in Japanese by using data from the Japanese Wikipedia site.

Description

• Embeddings are created by retrieving data from the Japanese Wikipedia site. Please refer to the Openapi site for details.

Getting Started

• Install the library with poetry.

$ poetry  install

• Create an env file and set the API key.

$ touch .env

# Edit env file
OPENAI_API_KEY=****************

• Creates a data set. When completed, a csv file of winter_olympics_2022_jp.csv will be saved.(It will take a few minutes to complete.)

$ poetry run python dataset.py

• Ask a question with the openapi API. First, the results of the question without embeddings are displayed. Next, the results using the embeddings you created are displayed.

$ poetry run python main.py

• It is displayed like this.The first answer is incorrect; the second is correct.

# Question
2022年の北京オリンピックで男子スノーボードハーフパイプの金メダルをもらった選手は誰ですか?

# Results of the question without embeddings
2022年の北京オリンピックはまだ開催されていないため、男子スノーボードハーフパイプの金メダリストはまだ決まっていません。

# Results using the embeddings
2022年の北京オリンピックで男子スノーボードハーフパイプの金メダルをもらった選手は平野歩夢です。