Guides

These pages focus on engineering decisions: where to integrate the kernels, how to measure them, and how to reason about FP8 trade-offs.

Integration Runtime contracts and model boundaries Choose between the functional API, module wrappers, and custom adapters. Performance Measurement and tuning Benchmark correctly, interpret metrics, and tune custom kernels. FP8 Quantization best practices Apply FP8 where it helps and keep numerically sensitive steps in higher precision.

---

Table of contents

• Integration Guide
• Performance Tuning
• FP8 Best Practices