# maderix/ANE

Training neural networks on Apple Neural Engine via reverse-engineered private APIs

Repository: https://github.com/maderix/ANE
Canonical: https://ross.abutalabs.com/products/ane
Language: Objective-C
License: MIT
License Family: permissive
Last push: 2026-03-10T10:21:05+00:00

## Health v2 (maintenance only)
Score: 47/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 71, release rhythm 35, longevity 13
- inputs: {"age_days": 186, "days_push": 176, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 7253, forks 960 (observed 2026-08-28T04:09:57.804255+00:00)

## What it is
A research project demonstrating backpropagation and neural network training directly on Apple's Neural Engine using reverse-engineered private APIs (_ANEClient, _ANECompiler), bypassing Apple's inference-only CoreML restriction. It includes benchmarks documenting ANE performance characteristics and serves as a reference for direct NPU access outside CoreML.

## Use cases
- train neural networks on Apple Neural Engine
- run backpropagation on ANE hardware
- benchmark Apple Neural Engine performance
- access ANE outside CoreML via private APIs
- research NPU training capabilities on Apple Silicon
- explore edge AI hardware optimization

## When to choose
- you're researching direct NPU access on Apple Silicon
- you want to benchmark ANE training performance
- you need a reference for reverse-engineering Apple private ML APIs
- you're exploring edge AI compiler optimization

## When to avoid
- you need a production training framework
- you want to train large models efficiently
- you need stable, maintained tooling with support
- you require high hardware utilization (currently ~5-9% of peak)
- you need a replacement for CoreML, MLX, or GPU training

## Facets
- artifact type: library
- maturity: experimental
- function: machine-learning, llm-training, reverse-engineering, gpu-computing, sdk
- domain: machine-learning, deep-learning, hardware, apple-ecosystem
- platform: cpp
- tags: apple-neural-engine, npu, backpropagation, private-apis, coreml, proof-of-concept, edge-ai, research, macos, objective-c

## Member repositories
- maderix/ANE (main) score 47

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:57.804255+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:39:43.291918+00:00, confidence not recorded.
  - readme: https://github.com/maderix/ANE (fetched 2026-08-28T04:09:57.804255+00:00, sha 9aa56598ad52)
- Data as of 2026-08-30T08:39:29.467469+00:00.