---
title: EasyDeL
emoji: 🔮
colorFrom: purple
colorTo: blue
sdk: static
pinned: false
---

<p align="center">
  <a href="https://github.com/erfanzar/EasyDeL">
    <img src="https://raw.githubusercontent.com/erfanzar/easydel/main/images/easydel-logo-with-text.png" height="80" alt="EasyDeL" />
  </a>
</p>

<p align="center">
  <a href="https://github.com/erfanzar/EasyDeL">
    <img src="https://img.shields.io/badge/GitHub-erfanzar%2FEasyDeL-111?logo=github&style=flat-square" alt="GitHub" />
  </a>
  <a href="https://pypi.org/project/easydel/">
    <img src="https://img.shields.io/pypi/v/easydel?style=flat-square" alt="PyPI" />
  </a>
  <a href="https://easydel.readthedocs.io/en/latest/">
    <img src="https://img.shields.io/badge/Docs-ReadTheDocs-1f72ff?logo=readthedocs&style=flat-square" alt="Docs" />
  </a>
  <a href="https://discord.gg/FCAMNqnGtt">
    <img src="https://img.shields.io/badge/Discord-Join-5865F2?logo=discord&style=flat-square" alt="Discord" />
  </a>
</p>

# EasyDeL

EasyDeL is an open-source framework for building, training, fine-tuning, converting, and serving modern ML models in **JAX** at scale. It is designed for people who want **the performance benefits of JAX** without giving up the **practical ergonomics** of the Hugging Face ecosystem.

## Purpose

JAX is extremely powerful, but scaling real training/inference workloads can still feel fragmented: model code, sharding, kernels, training loops, serving, and conversions often live in separate places. EasyDeL’s goal is to provide a cohesive toolkit where these pieces work together—while still staying readable and hackable.

## What EasyDeL focuses on

- **Scale-first**: multi-device training/inference across GPU/TPU with sharding-aware utilities.
- **Production inference**: a dedicated serving stack built for throughput and low latency.
- **Interoperability**: straightforward workflows with Hugging Face models and assets.
- **Hackability**: implementations you can actually read, debug, and modify.