SheetsTuner

SheetsTuner is a Google Sheets container-bound script that allows you to fine-tune LLMs (Large Language Models) on Vertex AI directly from your spreadsheet.

Overview

SheetsTuner simplifies the process of creating, managing, and uploading training datasets for fine-tuning language models. It leverages Google Sheets as an intuitive interface for preparing your data and Google Cloud Storage for storing the datasets.

Features

• Create training and validation datasets directly in Google Sheets
• Automatically split your data into training, validation, and test sets
• Export datasets in JSONL format required for fine-tuning
• Upload datasets to Google Cloud Storage buckets
• Call fine-tuned models directly from your spreadsheet

Built With

This project is built with Apps Script Engine, a modern development environment for Google Apps Script that enables the use of modern JavaScript features, modules, and build tools.

For a detailed explanation of the technique used in this project, check out this blog post: How Apps Script Became the Ultimate LLM Fine-Tuning Tool

Getting Started

Prerequisites

• Google account
• Google Cloud Platform project with enabled APIs:
  • Google Sheets API
  • Google Cloud Storage API
• Service account with appropriate permissions

Installation

1. Clone this repository
2. Install dependencies:

   npm i && npm run install:husky

Development

Running the local dev server with Vite

Run the Vite server and the Tailwind server in parallel in different terminal windows:

# Vite server
npm run dev

# Tailwind server
npm run build:css:watch

Formatting

The npm run format will run linting and pretty-printing with the --fix and --write options.

Managing Environments

This project is set up to be deployable to separate Google Apps Script projects acting as different environments. It works by copying relevant environmental files like .clasp.json from env-mgt/<environment_name> to the specified path (the root by default). Any number of files can be added to each environment. Configure the environment in the env-mgt/ENV.js file.

Building and deploying

To build the app for production, run:

npm run build:ui # to build the client-side
npm run build:gas # to build Google Apps Script library code

Or, if you want to build then push, run:

npm run build:push

The Apps Script code will be stored in a variable that executes an IIFE lib by default. Hence, to call an exported function from the bundled Apps Script code use lib.<functionName>(args) syntax.

Run tests

npm t

Git hooks

If you ran npm run install:husky, you will have configured a pre-commit git hook. Every time you commit your code, it will run formatting and testing by executing npm run format && npm t.

Author

👤 Dmitry Kostyuk

• Website: https://www.wurkspaces.dev
• Github: @WildH0g
• LinkedIn: @dmitrykostyuk
• StackOverflow: Dmitry Kostyuk

Show your support

Give a ⭐️ if this project helped you!