---
title: "BF16, FP8, INT8 & 4-Bit Inference"
description: "What each inference number format can hold: how bits are spent, the two FP8 encodings, INT8 and 4-bit with scales, reading the notation, and a 4-bit build."
canonical_url: "https://fanout.sh/inference-eng/curriculum/11-02"
md_url: "https://fanout.sh/inference-eng/curriculum/11-02.md"
access: "pro-preview"
---

# BF16, FP8, INT8 & 4-Bit Inference

What each inference number format can hold: how bits are spent, the two FP8 encodings, INT8 and 4-bit with scales, reading the notation, and a 4-bit build.

- Access: Fanout Pro (public preview)

## Public preview

The ladder in the previous lesson carried four names without opening any of them. Two formats can occupy the same byte and still disagree about which numbers they are able to hold, so a width is not quite the single quantity the ladder treated it as. This lesson opens the four the module is built on, then weighs a published…

## In this lesson

- How a format spends its bits
- The two FP8 encodings
- INT8, four bits, and the scale
- Reading the notation
- The weight of a published four-bit build

## Continue learning

Pro unlocks the complete lesson, video, practice, and code where available.

---
This representation contains public Fanout content only. Protected Pro lessons, account data, billing, checkout, and pricing are not included.

Browse the public content map: https://fanout.sh/sitemap.md