# mit-han-lab/proxylessnas

[ICLR 2019] ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware

Repository: https://github.com/mit-han-lab/proxylessnas
Canonical: https://ross.abutalabs.com/products/proxylessnas
Homepage: https://hanlab.mit.edu/projects/proxylessNAS/
Language: C++
License: MIT
License Family: permissive
Topics: automl, specialization, hardware-aware, acceleration, on-device-ai, efficient-model
Last push: 2024-08-30T23:51:21+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2832, "days_push": 733, "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 1447, forks 281 (observed 2026-08-28T04:04:45.412644+00:00)

## What it is
ProxylessNAS is a neural architecture search (NAS) framework that directly searches CNN architectures on the target task and target hardware (CPU, GPU, or mobile) without a proxy. It also provides pretrained, hardware-specialized image classification models loadable via PyTorch Hub.

## Use cases
- search neural network architectures optimized for a specific hardware platform
- get a pretrained efficient image classification model for mobile deployment
- reduce inference latency of CNNs on embedded devices
- run hardware-aware AutoML experiments in PyTorch
- benchmark latency-accuracy tradeoffs against MobileNetV2 and MnasNet

## When to choose
- you need architectures specialized for a particular device or latency budget
- you want efficient pretrained models for on-device image classification
- you are reproducing or building on the ProxylessNAS paper

## When to avoid
- you need actively maintained NAS tooling with broad framework support
- you are searching architectures for non-vision tasks
- you need a production AutoML pipeline rather than research code

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, llm-training
- domain: machine-learning, computer-vision, gpu-computing, artificial-intelligence
- platform: python, cpp, windows
- tags: neural-architecture-search, automl, hardware-aware-nas, efficient-inference, image-classification, on-device-ai, research-code, iclr-2019, gpu, linux, macos

## Member repositories
- mit-han-lab/proxylessnas (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:45.412644+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 2026-08-30T04:36:07.520760+00:00, confidence not recorded.
  - readme: https://github.com/mit-han-lab/proxylessnas (fetched 2026-08-28T04:04:45.412644+00:00, sha f81f78c04081)
- Data as of 2026-08-30T08:39:29.467469+00:00.