NOTE: Hands-On AI Software Training (3.5 hours) – See Prerequisites
Workshop will be held in AM and repeated in PM (Limit of 40 attendees each session)
Workshop Title: Building a GPU-Accelerated Retrieval Augmented Generation (RAG) Pipeline
Presenters: Nvidia Instructors (TBA)
Abstract:
Large language models (LLMs) show promising results in understanding and generating text. One limitation of the LLMs is that they are prone to hallucinate — to generate false knowledge. One reason for that is because they are agnostic to up-to-date and domain-specific data. Retrieval augmented generation (RAG) is a proposed solution for that issue, helping practitioners use current domain-specific data to augment LLM capabilities. Learn to create a RAG pipeline on GPU for the Q&A task, going over these steps:
— Understand the main concepts of LLMs and the RAG
— Pre-process the documents in the given document store
— Set up a vector database for retrieving relevant context from an embedding language model
— Combine the retrieved context with query + prompt and feed to an LLM to generate a response
Hands-on: Yes
Takeaways:
You will complete the first hours of Nvidia’s hands-on course, and receive credentials to then complete it on your own, leading to a Certificate of Competence. This may open opportunities for the mid-career SW/CS engineer.
Attendees:
Mid-career software developers (see prerequisites) Class limited to 40.
Equipment and software installed:
Bring your own laptop (cannot be corporate locked-down). Workshop includes working with Nvidia courseware and a dedicated Azure-based development environment.
Prerequisites:
The target audience is expected to have intermediate-level understanding of ML/DL and NLP pipelines. Basic knowledge of LLMs, [TensorFlow/Pytorch] and Python programming is required.