Skip to Main Content
Step-by-Step: Building a local inference engine on OCI

About This Workshop

Youtube Video

About This Workshop
Are you feeling like you've missed some steps in your AI journey? That you have a knowledge gap when it comes to GenAI? This workshop will walk you through every step in building an inference engine that can run offline while providing instant insights and knowledge to anything you need. You will learn how to install the required GPU drivers and CUDA toolkit as we prep the machine for the model. We will use Mistral 7B as the trained model and we will introduce the workshop in a fun learning scenario to keep it interesting!

Workshop Info

1 hour

"Welcome to the NovaSystems Team!
NovaSystems is a leader in deep space exploration. With dozens of satellites already launched to support their research initiatives, the team is now preparing for its most ambitious mission yet: the first crewed deep-space voyage to Mars.

As the spacecraft ventures farther from Earth, real-time communication with mission control will become increasingly limited. To overcome this challenge, NovaSystems has developed a locally running large language model (LLM) — a virtual mission expert designed to assist the crew by answering critical questions that would normally require input from ground control.

As NovaSystems' leading AI engineer, your task is to build and test a locally running large language model (LLM) capable of supporting the crew during deep-space operations. After evaluating various options, you’ve selected llama.cpp, paired with the latest Mistral 7B model, as the optimal solution due to its efficiency, local inference capability, and support for quantized models (which means lower-cost GPUs with les VRAM are a suitable option). 

Before deploying to physical flight hardware, you need to test this solution in a controlled environment. Since NovaSystems operates its infrastructure in Oracle Cloud Infrastructure (OCI) today, your testing will take place there.

 

Familiarity with OCI is helpful (this is not required, and we could alter the lab to include some things that are referenced like setting up a simple VCN). 

Familiarity with Linux (ssh, apt, etc). Again, not required as the lab tells you what to do. 

 

Other Workshops you might like

Other Workshops you might like Cards