A guide to building long-term compounding knowledge infrastructure. See details on GitHub .

AXLearn

An extensible deep learning library built on JAX/XLA, designed for developing, training and deploying large-scale models.

Introduction

AXLearn is an extensible library built on JAX and XLA to support development of large-scale deep learning models. It provides a configuration-based, modular approach to compose models and integrates with libraries like Flax and Hugging Face Transformers.

Key Features

  • Reusable model components and a declarative configuration system.
  • Support for large-scale distributed training using GSPMD-style global computation.
  • CLI and infra tooling for managing jobs, experiments and data.

Use Cases

  • Training large language and vision models with billions of parameters.
  • Running distributed training jobs on cloud or private clusters.
  • Serving as a research-to-production framework for model development and baselines.

Technical Details

  • Built on JAX/XLA for efficient compilation and execution.
  • Modular configuration for reproducibility and experiment management.
  • In-repo docs (docs/) provide guidance for getting started, concepts and CLI usage.

Comments

AXLearn
Resource Info
Author Apple
Added Date 2025-09-30
Open Source Since 2023-02-25
Tags
Open Source Framework ML Platform