NKI Kernels

This tutorial guides you through writing custom high-performance kernels for Trainium using the Neuron Kernel Interface (NKI).

---

Overview

NKI (Neuron Kernel Interface) is a low-level programming interface that allows you to directly leverage Neuron device hardware capabilities to write custom compute kernels.

Use Cases

• Custom attention mechanisms
• Fused operations
• Special activation functions
• Quantization kernels

---

Prerequisites

• Neuron SDK 2.18 or higher
• Basic NKI programming knowledge
• trn1 instance

---

Step 1: NKI Basics

Coming Soon

Detailed step-by-step guide will be added soon.

---

References

• NKI Programming Guide
• NKI Samples