OpenAI Dataset Generator & Validator

A Streamlit application for creating, validating, and managing fine-tuning datasets for OpenAI models.

Features

• Create JSONL format datasets for fine-tuning
• Validate dataset format and structure
• Upload datasets directly to OpenAI
• Create and monitor fine-tuning jobs
• Track fine-tuned model status
• Analyze token distribution and dataset statistics

Prerequisites

• Python 3.7+
• OpenAI API key
• Streamlit
• Required Python packages (see Installation)

Installation

1. Clone the repository:

git clone <repository-url>
cd openai-dataset-generator

2. Install required packages:

pip install streamlit openai python-dotenv tiktoken numpy

3. Create a .env file in the root directory and add your OpenAI API key:

OPENAI_API_KEY=your_api_key_here

Usage

1. Start the Streamlit application:

streamlit run app.py

2. Follow the step-by-step process in the web interface:

   • Step 1: Create a fine-tuning dataset

     • Add a system message
     • Provide training examples (minimum 10)
     • Generate and download JSONL file

   • Step 2: Verify data format and structure

     • Check for format errors
     • Review token distribution
     • Analyze conversation lengths

   • Step 3: Upload dataset to OpenAI

     • Automatically uploads JSONL file
     • Receives file ID for tracking

   • Step 4: Create fine-tuning job

     • Initialize training process
     • Uses gpt-3.5-turbo model

   • Step 5: Monitor fine-tuned model

     • Track training progress
     • View model statistics and parameters
     • Get fine-tuned model identifier

Dataset Validation Checks

The validator performs comprehensive checks including:

• Data format and structure validation
• Token count analysis
• Message distribution statistics
• System and user message presence
• Token limit compliance (16,385 tokens per example)
• Pricing estimation based on token usage