Efficient Large Language Model (LLM) Customization
Contact IT Dojo for current pricing, available dates, and a custom quote tailored to your team or organization.
Course Duration
1 Day
Audience
Employees of federal, state and local governments; and businesses working with the government.
Prerequisites
Basic Python programming experience and familiarity with machine learning concepts recommended.
Course Description
Out-of-the-box pretrained large language models rarely fit a specific use case perfectly. This one-day course goes beyond using off-the-shelf LLMs and teaches a variety of techniques to efficiently customize pretrained models for specific use cases without engaging in the full computational cost of training a model from scratch. Participants work hands-on with parameter-efficient customization approaches and evaluate the trade-offs between them.
Learning Objectives
- Choose between prompting, retrieval augmentation, and fine-tuning for a given use case
- Explain parameter-efficient fine-tuning approaches such as LoRA
- Build a simple retrieval-augmented pipeline to ground model output
- Evaluate the cost, complexity, and output quality trade-offs of different customization methods
Course Outline
1. Why and When to Customize an LLM
- Prompting versus retrieval augmentation versus fine-tuning: choosing the right approach
- Cost and complexity trade-offs of each approach
- Evaluating whether customization is really needed
2. Parameter-Efficient Customization Techniques
- Overview of parameter-efficient fine-tuning approaches such as LoRA
- Adapters and lightweight customization methods
- Hardware and cost considerations
3. Retrieval-Augmented Approaches
- Grounding model output in your own data without retraining
- Building and evaluating a simple retrieval pipeline
- Combining retrieval with lightweight fine-tuning
4. Hands-On Lab
- Customizing a pretrained model for a sample use case
- Evaluating output quality before and after customization
- Discussing deployment considerations
5. Wrap-Up
- Choosing the right customization approach for your own use case
- Final questions and key takeaways
Frequently Asked Questions
What does this course cover?
This course covers efficient techniques for customizing pretrained large language models, including parameter-efficient fine-tuning and retrieval-augmented approaches.
How long is this course?
Efficient Large Language Model (LLM) Customization is a one-day, hands-on course, available as live remote online instruction or on-site at your facility.
Who should attend?
Developers, data scientists, and ML engineers who want to customize pretrained LLMs for specific use cases.
Does IT Dojo offer this training on-site at government or DoD facilities?
Yes. IT Dojo delivers this course on-site at government agencies, DoD commands, and contractor facilities, in addition to live online delivery.
How do I register for this course?
IT Dojo training is employer sponsored. Contact IT Dojo via the Request Training form or call 757-216-3656 to schedule this course for your team.