# Baseten — What makes a high-quality NVFP4 quantization?

- Company: Baseten (baseten.co)
- Announced: 2026-10-09T14:16:52+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.baseten.co/blog/what-makes-a-high-quality-nvfp4-quantization
- Record: https://forck.live/items/21695-what-makes-a-high-quality-nvfp4-quantization
- Subject: Baseten

Baseten's post explains three approaches to assigning precision formats per layer in NVFP4 quantization: heuristic-based algorithms, isolated-layer sensitivity scoring, and SaturationQuant, which incorporates saturation so scores can be summed across a mixed-precision recipe. The post also covers calibration steps and the role of block scales.

## Evidence

Verbatim from https://www.baseten.co/blog/what-makes-a-high-quality-nvfp4-quantization:

> A high-quality NVFP4 quantization comes down to choosing which layers can run at 4 bits and which need more precision. This post covers three ways to assign a format to each layer: heuristic-based algorithms decide by architecture; isolated-layer sensitivity scoring measures one layer at a time, so its scores can't be added; and SaturationQuant builds saturation into each score, so they can be added across a mixed-precision recipe.

---

Record: https://forck.live/items/21695-what-makes-a-high-quality-nvfp4-quantization
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md