Fine-Tuning Models on Amazon Bedrock: Step-by-Step Implementation Guide

Fine-tuning foundation models on Amazon Bedrock allows you to customize AI models for your specific use case. In this hands-on tutorial, we’ll walk through the complete process of fine-tuning a model on Amazon Bedrock, from data preparation to deployment.

What We’ll Build In This Amazon Bedrock Fine-Tuning Tutorial

By the end of this tutorial, you’ll have:

A fine-tuned Amazon Titan model customized for your domain
Complete understanding of the fine-tuning workflow
Production-ready code for your own implementations

Prerequisites

Before going through the steps, ensure to have the following:

AWS account with Bedrock access enabled
AWS CLI configured with appropriate permissions
Python 3.8+ with boto3 installed
Basic understanding of machine learning concepts

Steps To Fine-Tune AI Models on Amazon Bedrock

Step 1: Enable Amazon Bedrock Access

First, we need to request access to Amazon Bedrock models.

Navigate to the Amazon Bedrock console
Go to “Model access” in the left sidebar
Click “Manage model access.”

Select the models you want to use (for this tutorial, select “Amazon Titan Text G1 – Express”)
Click “Request model access.”

Note: Model access approval can take a few minutes to several hours, depending on the model.

Step 2: Set Up IAM Permissions

Create an IAM role that Bedrock can use for fine-tuning operations.

Go to the IAM console.
Click “Roles” → “Create role.”
Attach the following policies:
- AmazonBedrockFullAccess
- AmazonS3FullAccess (for accessing training data)
Name your role bedrock-full-access.
Note the Role ARN – we’ll use it later.

Step 3: Prepare Your Training Data

For this example, we’ll create a customer service chatbot training dataset.

Create Training Data

Create a file named training_data.jsonl with the following format:

{"prompt": "How do I reset my password?", "completion": "To reset your password: 1. Go to the login page 2. Click 'Forgot Password' 3. Enter your email address 4. Check your email for reset instructions 5. Follow the link to create a new password"}
{"prompt": "What are your business hours?", "completion": "Our customer service is available Monday through Friday, 9 AM to 6 PM EST. For urgent issues outside these hours, please email support@company.com"}
{"prompt": "How do I cancel my subscription?", "completion": "To cancel your subscription: 1. Log into your account 2. Go to Account Settings 3. Click on Subscription 4. Select 'Cancel Subscription' 5. Confirm cancellation. You'll retain access until your billing period ends"}
{"prompt": "Where can I track my order?", "completion": "You can track your order by: 1. Logging into your account 2. Going to 'My Orders' 3. Clicking on the order number 4. Viewing tracking information. You'll also receive tracking updates via email"}
{"prompt": "Do you offer refunds?", "completion": "Yes, we offer refunds within 30 days of purchase. To request a refund: 1. Contact our support team 2. Provide your order number 3. Explain the reason for refund 4. We'll process it within 3-5 business days"}

Validation Data (Optional)

Create validation_data.jsonl for model evaluation:

{"prompt": "How long does shipping take?", "completion": "Standard shipping takes 5-7 business days. Express shipping takes 2-3 business days. Free shipping is available on orders over $50"}
{"prompt": "Can I change my order after placing it?", "completion": "You can modify your order within 1 hour of placing it by contacting customer service. After that, the order enters processing and cannot be changed"}

Data Quality Tips

For better results:

Consistent Format: Keep prompt/completion structure uniform
Representative Examples: Cover all scenarios your model will encounter
Quality over Quantity: 100 high-quality examples often outperform 1000 poor ones
Length Balance: Mix short and detailed responses

Step 4: Upload Data to S3

Create an S3 bucket and upload your training data:

# Create S3 bucket
aws s3 mb s3://bedrock-finetuning-bucket

# Upload training data
aws s3 cp training_data.jsonl s3://bedrock-finetuning-bucket/data/training_data.jsonl

# Upload validation data (optional)
aws s3 cp validation_data.jsonl s3://bedrock-finetuning-bucket/data/validation_data.jsonl

Step 5: Create the Fine-Tuning Job

Now we’ll create the fine-tuning job using the AWS Console.

5.1 Using AWS Console

Navigate to the Amazon Bedrock console.
Click “Custom models” in the left sidebar.
Click “Create custom model.”

Fill in the model details:
- Model name: customer-service-chatbot-v1
- Base model: Amazon Titan Text G1 - Express
- Customization method: Fine-tuning

Configure training settings:
- Training data location: s3://bedrock-finetuning-bucket/data/training_data.jsonl
- Validation data location: s3://bedrock-finetuning-bucket/data/validation_data.jsonl
- Output data location: s3://bedrock-finetuning-bucket/output/

Set hyperparameters:
- Learning rate: 0.0001
- Batch size: 1
- Epochs: 3

Select the IAM role we created earlier: bedrock-full-access.
Click “Create custom model.”

5.2 Using Python SDK (Alternative)

import boto3
import json

# Initialize Bedrock client
bedrock = boto3.client('bedrock', region_name='us-east-1')

# Create fine-tuning job
response = bedrock.create_model_customization_job(
    jobName='customer-service-chatbot-job',
    customModelName='customer-service-chatbot-v1',
    roleArn='arn:aws:iam::123456789012:role/BedrockFineTuningRole',
    baseModelIdentifier='amazon.titan-text-express-v1',
    trainingDataConfig={
        'S3Uri': 's3://bedrock-finetuning-bucket/data/training_data.jsonl'
    },
    validationDataConfig={
        'S3Uri': 's3://bedrock-finetuning-bucket/data/validation_data.jsonl'
    },
    outputDataConfig={
        'S3Uri': 's3://bedrock-finetuning-bucket/output/'
    },
    hyperParameters={
        'epochCount': '3',
        'batchSize': '1', 
        'learningRate': '0.0001'
    },
    tags={
        'Environment': 'Development',
        'Project': 'Customer-Service-Bot'
    }
)

print(f"Fine-tuning job created: {response['jobArn']}")

Step 6: Monitor Training Progress

Go to “Custom models” in the Bedrock console.
Find your model and click on it.

Step 7: Test Your Fine-Tuned Model

Once training completes, you can test your model.

Get Model ARN

In the Bedrock console, go to “Custom models.”
Find your completed model.
Copy the Model ARN.

Test Using AWS Console

Go to the “Text” playground in the Bedrock console
Select your custom model from the dropdown
Enter a test prompt: “How do I reset my password?”
Click “Run” and observe the response

Conclusion

This guide provides an end-to-end approach to fine-tuning foundation models on Amazon Bedrock, enabling organizations to customize AI models for specific business or research use cases.

Key Takeaways:

Prioritize data quality: High-quality datasets are essential for effective fine-tuning.
Start with small experiments: Validate approaches before scaling up.
Monitor model performance: Track training metrics and adjust as needed.
Optimize costs: Manage compute resources and experiment strategically to control expenses.
Plan for production deployment: Ensure models are monitored and maintained in real-world scenarios.

The code repository shared in this guide serves as a reusable framework for future fine-tuning projects, allowing experimentation with different models, datasets, and hyperparameters to achieve highly specialized, production-ready AI models using Amazon Bedrock.