AI Engineering: Customizing LLMs for Business (Fine-Tuning LLMs with QLoRA & AWS)

7h 12m 10s
English
Paid

Master the in-demand skill that companies are actively seeking: developing and implementing custom Large Language Models (LLMs). In this course, you will learn how to fine-tune open LLMs using corporate data and deploy your models efficiently with AWS tools like SageMaker, Lambda, and API Gateway, alongside creating intuitive interfaces with Streamlit for employees and clients.

Course Overview

This is not "just another introductory AI course." Instead, it is a practical, in-depth exploration of the skills that distinguish AI engineers on real-world projects. You'll engage in fine-tuning using QLoRA—an approach that significantly minimizes resource consumption—and transition the model into a production-ready service.

Key Learning Outcomes

What you will master:

  1. Fine-tuning open-source LLMs with your own datasets, including corporate data.
  2. Hands-on experience with QLoRA, bfloat16 training, dataset chunking, and attention masks.
  3. Utilizing the Hugging Face ecosystem, including the Estimator API, and setting up an MLOps pipeline on AWS.
  4. Model deployment and integration with tools like SageMaker endpoints, Lambda, API Gateway, and monitoring.
  5. Creation of a simple business UI using Streamlit.

Course Outcomes

From theory to code to production—experience the complete development cycle of applied AI tailored for business scenarios.

Target Audience and Career Paths

This course is designed to benefit and prepare professionals for roles such as:

  1. AI Engineer / ML Engineer: designing, fine-tuning, and producing models.
  2. AI Specialist: developing applied AI solutions.
  3. Data Scientist: preparing data, performing EDA, and building models for organizational tasks.
  4. AI Research Scientist: focusing on attention mechanisms and in-depth LLM work.
  5. Cloud Engineer: crafting architecture and implementing best deployment practices on AWS.
  6. DevOps Engineer: automating, releasing, and monitoring ML services using tools like CloudWatch.
  7. Software Engineer: integrating scalable models into applications.
  8. Data Engineer: constructing data pipelines, managing storage (e.g., S3), and preprocessing.
  9. Technical Product Manager: planning and releasing ML products, focusing on metrics and monitoring.

If you're looking to harness the "AI wave," customizing LLMs for business needs offers a substantial entry point and growth opportunity.

About the Author: zerotomastery.io

zerotomastery.io thumbnail
Whether you are just starting to learn to code or want to advance your skills, Zero To Mastery Academy will teach you React, Javascript, Python, CSS and more to help you advance your career, get hired and succeed at some of the top companies in the world.

Watch Online 58 lessons

This is a demo lesson (10:00 remaining)

You can watch up to 10 minutes for free. Subscribe to unlock all 58 lessons in this course and access 10,000+ hours of premium content across all courses.

View Pricing
0:00
/
#1: Course Introduction (What We're Building)
All Course Lessons (58)
#Lesson TitleDurationAccess
1
Course Introduction (What We're Building) Demo
05:20
2
Signing in to AWS
04:31
3
Creating an IAM User
05:30
4
Using our new IAM User
03:13
5
What To Do In Case You Get Hacked!
01:31
6
Creating a SageMaker Domain
02:29
7
Logging in to our SageMaker Environment
04:54
8
Introduction to JupyterLab
07:38
9
Sagemaker Sessions, Regions, and IAM Roles
07:51
10
Examining Our Dataset from HuggingFace
13:30
11
Tokenization and Word Embeddings
09:09
12
HuggingFace Authentication with Sagemaker
04:22
13
Applying the Templating Function to our Dataset
08:44
14
Attention Masks and Padding
15:56
15
Star Unpacking with Python
04:04
16
Chain Iterator, List Constructor and Attention Mask example with Python
10:23
17
Understanding Batching
08:12
18
Slicing and Chunking our Dataset
07:32
19
Creating our Custom Chunking Function
16:07
20
Tokenizing our Dataset
09:31
21
Running our Chunking Function
04:31
22
Understanding the Entire Chunking Process
08:33
23
Uploading the Training Data to AWS S3
05:54
24
Setting Up Hyperparameters for the Training Job
06:48
25
Creating our HuggingFace Estimator in Sagemaker
06:46
26
Introduction to Low-rank adaptation (LoRA)
08:12
27
LoRA Numerical Example
10:56
28
LoRA Summarization and Cost Saving Calculation
09:09
29
(Optional) Matrix Multiplication Refresher
04:46
30
Understanding LoRA Programatically Part 1
12:33
31
Understanding LoRA Programatically Part 2
05:49
32
Bfloat16 vs Float32
08:11
33
Comparing Bfloat16 Vs Float32 Programatically
06:33
34
Setting up Imports and Libraries for the Train Script
07:20
35
Argument Parsing Function Part 1
07:57
36
Argument Parsing Function Part 2
10:55
37
Understanding Trainable Parameters Caveats
14:31
38
Introduction to Quantization
07:36
39
Identifying Trainable Layers for LoRA
07:20
40
Setting up Parameter Efficient Fine Tuning
04:36
41
Implement LoRA Configuration and Mixed Precision Training
10:35
42
Understanding Double Quantization
04:22
43
Creating the Training Function Part 1
14:15
44
Creating the Training Function Part 2
07:17
45
Exercise: Imposter Syndrome
02:57
46
Finishing our Sagemaker Script
05:09
47
Gaining Access to Powerful GPUs with AWS Quotas
05:11
48
Final Fixes Before Training
03:55
49
Starting our Training Job
07:16
50
Inspecting the Results of our Training Job and Monitoring with Cloudwatch
11:24
51
Deploying our LLM to a Sagemaker Endpoint
17:58
52
Testing our LLM in Sagemaker Locally
08:19
53
Creating the Lambda Function to Invoke our Endpoint
08:56
54
Creating API Gateway to Deploy the Model Through the Internet
02:37
55
Implementing our Streamlit App
05:12
56
Streamlit App Correction
03:27
57
Congratulations and Cleaning up AWS Resources
02:39
58
Thank You!
01:18
Unlock unlimited learning

Get instant access to all 57 lessons in this course, plus thousands of other premium courses. One subscription, unlimited knowledge.

Learn more about subscription