Getting Started with Fine Tuning

This comprehensive guide will walk you through the essential steps to begin your fine tuning journey in Serenity* AI Hub. By the end of this guide, you'll have successfully created your first fine tuning request and understand the key concepts needed to customize AI models for your specific needs.

Prerequisites

Before you begin, ensure you have:

• ✅ An active Serenity* AI Hub subscription with fine tuning access
• ✅ A dataset prepared in JSONL format (see Data Preparation Basics below)
• ✅ Basic understanding of your use case and desired model behavior

Step 1: Access the Fine Tuning Dashboard

If your subscription includes fine tuning access, you'll find the fine tuning option in the sidebar navigation.

Click on the fine tuning option to access the dashboard:

tip

If you don't see the fine tuning option in your sidebar, contact your account administrator to verify your subscription includes fine tuning capabilities.

Step 2: Create Your First Fine Tuning Request

To create a new fine tuning request, click the "New Fine Tuning" button from the dashboard.

Required Fields

You'll need to complete three essential fields:

1. Name: A descriptive identifier for your fine tuning request

   • Choose a name that clearly describes your use case (e.g., "Customer Support Bot", "Technical Documentation Writer")
   • This will help you identify the model later in your dashboard

2. Model: The base model you want to fine-tune

   • Not all models support fine tuning
   • Popular options include GPT-3.5-turbo and GPT-4 variants
   • Consider your performance needs and budget when selecting

3. Dataset: Your training data in JSONL format

   • Must be a valid JSONL file (newline-delimited JSON)
   • See the Data Preparation Basics section for formatting requirements

Creating Your Request

1. Fill in all required fields
2. Review your selections carefully
3. Click the "Create" button to submit your request

Important

Fine tuning requests cannot be modified once submitted. Double-check your dataset and model selection before creating the request.

After creation, the new fine tuning request will appear in the index grid and the first status will be "Pending", once the the process starts it changes to "Validating files" (see Understanding Fine Tuning Request Statuses below).

Once, the process is complete, the status will be "Success". At this point, you can also see the fine-tuning tokens that were used during the training process and the model code that you'll need to reference when using your fine-tuned model.

And the details will show more data, like the cost of the fine tuning process

Using your fine tuning model is straight forward, just choose it in the model selection

Understanding Fine Tuning Request Statuses

Once you've created your fine tuning request, it will progress through several status stages. Understanding these statuses helps you track progress and identify any issues that may arise.

Status Overview

Your fine tuning request can have the following statuses:

Status	Description	What's Happening
Pending	Initial state after creation	Request is being prepared for processing
Validating Files	File validation in progress	System is checking your dataset format and content
Queued	Waiting for processing resources	Request is valid and waiting in the training queue
Running	Training in progress	Model is actively being fine-tuned with your data
Success	Training completed successfully	Your fine-tuned model is ready to use
Error	Training failed	An issue occurred during the process
Cancelled	Request was cancelled	Training was stopped by user or system
Disabled	Request has been disabled	Model access has been restricted

What to Expect During Each Stage

Pending → Validating Files

• Duration: Usually 1-2 minutes
• What happens: System validates your JSONL file format and content
• Action needed: None, wait for validation to complete

Validating Files → Queued

• Duration: Immediate if validation passes
• What happens: Your request enters the training queue
• Action needed: None, your request will be processed in order

Queued → Running

• Duration: Varies based on queue length
• What happens: Training resources are allocated and training begins
• Action needed: None, training will start automatically

Running → Success

• Duration: Depends on dataset size
• What happens: Your model is being fine-tuned and validated
• Action needed: None, you can monitor progress in the dashboard

Something went wrong?

If your request shows an Error status enter the details of your request and an Errors part will be visible.

Monitoring Progress

You can refresh the dashboard to see real-time status updates. The system automatically updates the status as your request progresses through each stage.

Data Preparation Basics

Proper data preparation is crucial for successful fine tuning. Your dataset must be in JSONL (JSON Lines) format, where each line contains a complete JSON object representing a training example.

JSONL Format Requirements

Each line in your JSONL file should follow this structure:

{"messages": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "What is the capital of France?"}, {"role": "assistant", "content": "The capital of France is Paris."}]}

Data Structure Guidelines

• System Message: Define the model's behavior and role
• User Message: The input or query from the user
• Assistant Message: The desired response from the model

Example Dataset

Here's a sample of well-formatted training data for a customer support bot:

{"messages": [{"role": "system", "content": "You are a customer support representative for TechCorp. Be helpful, professional, and concise."}, {"role": "user", "content": "How do I reset my password?"}, {"role": "assistant", "content": "To reset your password, visit our login page and click 'Forgot Password'. Enter your email address, and we'll send you a reset link within 5 minutes."}]}
{"messages": [{"role": "system", "content": "You are a customer support representative for TechCorp. Be helpful, professional, and concise."}, {"role": "user", "content": "What are your business hours?"}, {"role": "assistant", "content": "Our customer support is available Monday through Friday, 9 AM to 6 PM EST. For urgent issues outside these hours, please email [email protected]."}]}

Quality Guidelines

• Minimum Examples: Include at least 50-100 high-quality examples
• Consistency: Maintain consistent formatting and response style
• Diversity: Cover various scenarios your model will encounter
• Accuracy: Ensure all responses are factually correct and appropriate

Pro Tip

Start with a smaller, high-quality dataset (50-100 examples) for your first fine tuning experiment. You can always create additional fine-tuned models with larger datasets once you've validated your approach.

note

Training times are estimates and may vary based on model complexity and system load.