Scaling GenAI: Experimentation to Transformation

• Key capabilities for operationalizing GenAI models at scale
• Success factors for GenAI use cases
• Common challenges and how to avoid them
• Inferencing GenAI models cost-effectively

• The state of GenAI: where GenAI is delivering and missing expectations
• The challenges: the real risks and remaining barriers to impact
• The future: what advances are underway and what can we expect over the next year

We tackle the biggest misconceptions around driving business value with GenAI, including...

• Generative AI is the most important kind of AI
• Bigger is better
• You don’t need data scientists for Generative AI
• I need to hire prompt engineers
• LLMOps is a separate discipline from MLOps
• Someone else is responsible for responsible AI

• The AI and ML challenges in the world of manufacturing
• The benefit of data-centric approaches to AI
• The relationship between speed, risk, and process change
• Best practices for responsible and impactful AI projects

• Review the motivation and theory behind PEFT (parameter-efficient fine-tuning) techniques
• Discover the power of quantization with the Hugging Face Trainer on Domino using Falcon-40b
• Investigate LoRA with the Falcon-7b LLM using PyTorch Lightning

• Learn about ZeRO and how it helps you fine-tune models across GPUs
• Using DeepSpeed and Ray to fine-tune GPTJ-6b model
• Load the model for inference onto a GPU or a CPU

• Understand the structure of prompts.
• Help LLM understand what you need using zero-shot, one-shot, and few-shot in-context training.
• Work faster by reusing your prompts and leveraging open-source prompt template libraries.
• Explore automation techniques such as prompt chaining.

• Theoretical Foundations: Natural Language Processing (NLP) basics and retrieval techniques.
• How RAG Works: RAG interactions in detail with real-world case studies.
• Applications and Challenges: Examine RAG-powered chatbots and search engines, and review RAG's limits.
• Demo & Q&A: We'll show you how RAG works in Domino and leave you time for Q&A at the end of the session.

• The benefits and challenges of working with multiple LLM providers
• How Jupyter AI in Domino with your choice of LLM coder models to accelerate development (and documentation, too!)
• Why harnessing Domino's new AI Gateway can simplify the transition to best-of-breed LLM solutions.

• Learn how to leverage the NVIDIA NeMo Framework to accelerate large language model (LLM) training by up to 30%.
• Explore using NeMo Megatron, a powerful transformer, for fine-tuning on Domino's Enterprise AI Platform.
• Cover the entire model life cycle, from setup in Domino's AI Project Hub to customizing environments for fine-tuning and utilizing NeMo Megatron for text encoding and virtual token generation.