y-temp4 / refiner

The open-source Refraction alternative.

Refiner

The open-source Refraction alternative.

https://refiner.dev

Build with

• OpenAI GPT3 API
• Next.js
• tRPC
• NextAuth.js
• Prisma
• Mantine
• Tailwind CSS

Development

Prerequisites:

• Node.js
• Docker
• Yarn

# Clone the repo
git clone https://github.com/y-temp4/refiner

# Go to the project folder
cd refiner

# Install packages
yarn

# Set up .env files
# Please setup github keys and OpenAPI keys
cp .env.sample .env

# Set up DB
docker compose up
yarn prisma db push

# Start dev server
yarn dev

Then you can access http://localhost:3000.

refiner's Issues

How to correctly estimate the number of tokens

In this project, we are counting the token numbers of the input code using the encode function from @nem035/gpt-3-encoder.

However, this value may differ from the actual token number output when using the gpt-3.5-turbo API.

If anyone knows a better way to count tokens accurately, please let me know.