---
title: "Discussion on \"High Level Neural Network Modeling with Flax\" | Hashnode"
description: "Discussion on \"High Level Neural Network Modeling with Flax\". In our previous article, we broke down why JAX requires explicit state handling. We saw that wrapping our parameters, optimizer momentum, and PRNG keys into a clean carry pattern lets us use jax.jit,"
url: "https://hashnode.com/posts/high-level-neural-network-modeling-with-flax/6a8626026572584351d689f0"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [High Level Neural Network Modeling with Flax](https://marshud.dev/high-level-neural-network-modeling-with-flax?utm_source=hashnode&utm_medium=feed)

In our previous article, we broke down why JAX requires explicit state handling. We saw that wrapping our parameters, optimizer momentum, and PRNG keys into a clean carry pattern lets us use jax.jit,

Aug 19 · 7 min read · marshud.dev  
Read the article

## Written by

[Marshud](https://hashnode.com/@marshud256)

@marshud256 · localhost

  
[View profile](https://hashnode.com/@marshud256)

Learning more tech

Software Engineer

## Responses 1

Markdown supported · ⌘↵ to post

[Julian Neagu](https://hashnode.com/@julianneagulab)

[Aug 20](https://hashnode.com/posts/high-level-neural-network-modeling-with-flax/6a8626026572584351d689f0/comment/6a86f75232254c1d143eedf1)

The `init` and `apply` split is probably the part that clicked for me most. Keeping model structure separate from the actual weights feels much cleaner. Especially when you start adding dropout and batch stats.
