Key Takeaways
- Foundational Concepts: Understand the core principles of transformers architecture, their evolution, and their application in building enterprise AI systems.
- Generative AI Frameworks: Learn and apply strategies for building enterprise AI applications using generative AI software frameworks (e.g., langchain, llamaindex). Fine-tune pre-trained models with available foundational models for specific enterprise tasks, including zero-shot learning on computer vision and retrieval-augmented generation.
- Performance Optimization: Explore methods to enhance enterprise AI applications using reward-based reinforcement learning for performance optimization.
- Deployment Strategies: Navigate challenges and strategies for deploying enterprise AI systems in production environments. Optimize models for efficient training and inference, with an understanding of model and data parallelism.
- Practical Application: Apply theoretical knowledge through practical labs, building and deploying enterprise AI applications in real-world scenarios.
Our course participants will learn to first fine-tune real world enterprise applications with LLMs using foundational models from the closed source ones such as GPT 4o from Open AI and Claude 3.5 from Anthropic and open source ones such as lLAMA 3 from Meta or Hugging Faces and proceed to build their own LLMs if the foundational models are not able to fulfil the need.
Participants will benefit from a careful balance of lectures and practical workshops. There will be projects and assessment to reinforce participants’ learning as part of the course.
Who Should Attend
This course is suitable for information technology professionals who are planning to build their own enterprise AI applications with either with their own LLMs or fine-tune LLMs with foundational models.
This course will be useful for CTOs and technical leaders, data engineers, data scientists ML engineers, and software developers advancing in Large Language Models (LLMs) for fine-tuning, deployment, and training.
Pre-requisites
This is an intensive intermediate course.
• Participants should have intermediate mathematics and statistics knowledge, e.g. calculating boolean algebra (logic), and probability.
• Participants should have intermediate computer literacy and software engineering fundamentals, e.g. using Windows or Linux or MacOS, Microsoft Office or LibreOffice, VMware or VirtualBox, and aware of web application, and client-server software architecture.
• Participants should have current or prior hands-on coding experience in one or more high-level computer programming languages, preferable in Java. Experiences with Python, R, or structured query language (SQL) would have added advantages.
• Participants without programming experience should self-study basic Java or Python.
• Knowledge of deploying applications on the cloud such as AWS and GCP are a plus.
What Will be Covered
- Introduction to Large Language Models (LLMs)
- LLM Pre-Training and Scaling Laws
- Building with a Foundational Model with Langchain
- Fine-tuning LLMs with Instruction
- Parametric Effective Fine-Tuning (PEFT)
- Fine-tuning a Generative AI model for dialogue summarization – Lab Hands On
- Reinforcement Learning in LLMs with Human Feedback, Reward Hacking and Scaling
- Applying Reinforcement Learning into LLM – Lab Hands On
- Building your own LLM and choice of architecture
- Planning the LLM Model pre-training
- Gathering, Selection & Pre-processing dataset for LLM
- Tokenization
- Hyper-parameter tuning
- Evaluation & Finetuning of Pre-trained LLMs
- Scaling of LLMs
- Retrieval Augmented Generation (RAG) Implementations
- Responsible AI, LLMs reasoning and plan with chain of thought
- In-class Project review
Fees & Subsidies
Certificate
The ISS Certificate of Completion will be issued to participants who have attended at least 75% of the course and pass the required assessments.
Preparing for Your Course
NUS-ISS Course Registration Terms and Conditions
Find out more.
NUS-ISS and Learner’s Commitment and Responsibilities
Find out more.
WIFI Access
WIFI access will be made available to participants.
Venue
NUS-ISS
25 Heng Mui Keng Terrace
Singapore 119615
Click HERE for directions to NUS-ISS
In the event of a change of venue, participants are advised to refer to the acceptance email sent one week prior to the commencement date.
Course Confirmation
All classes are subject to confirmation and NUS-ISS will send an acceptance email to participants one week prior to the commencement date. Confirmed registrants are to attend and complete all lectures, class exercises, workshops and assessments (where applicable). Additionally, all responses to feedbacks and surveys conducted by NUS-ISS and its partners must be submitted. All training and assessments will be delivered as described in the course webpage.
General Enquiry
Please feel free to write to ask-iss@nus.edu.sg if you have any enquiry or feedback.