# pytorch/executorch

On-device AI across mobile, embedded and edge for PyTorch

Repository: https://github.com/pytorch/executorch
Canonical: https://ross.abutalabs.com/products/executorch
Homepage: https://executorch.ai
Language: Python
License: NOASSERTION
License Family: other
Topics: deep-learning, embedded, machine-learning, mobile, neural-network, tensor, gpu
Last push: 2026-08-26T23:13:55+00:00

## Health v2 (maintenance only)
Score: 95/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 86, longevity 100
- inputs: {"age_days": 1650, "days_push": 7, "days_rel": 19, "gap_med": 65.5, "n_releases_24m": 11}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4953, forks 1122 (observed 2026-08-28T04:09:03.701893+00:00)

## What it is
ExecuTorch is PyTorch's framework for exporting and running AI models on-device across mobile, embedded, and edge hardware, with a tiny (~50KB) runtime and 12+ hardware backends. It supports LLMs, vision, speech, and multimodal models with native PyTorch export and no intermediate format conversions.

## Use cases
- run llm inference on a smartphone offline
- deploy pytorch models to mobile without onnx conversion
- run computer vision models on embedded devices
- on-device speech recognition with privacy
- deploy ai models to microcontrollers with tiny runtime
- quantize and optimize models for edge hardware
- run multimodal ai on ar/vr headsets

## When to choose
- you need privacy-preserving, offline inference on phones or embedded devices
- your models are already in PyTorch and you want direct export without format conversion
- you need a very small runtime footprint down to microcontroller scale
- you need to target many hardware backends (CPU, GPU, NPU) from one toolchain

## When to avoid
- your inference runs server-side with abundant compute and no edge constraints
- your models are in TensorFlow/JAX and you prefer TFLite or ONNX ecosystems
- you need training on device rather than inference
- you want a turnkey app rather than an SDK to integrate into your own code

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-inference, deep-learning, compiler, sdk
- domain: machine-learning, deep-learning, large-language-models, embedded-systems, mobile-development, computer-vision, speech-processing
- platform: embedded, python, cpp, cross-platform
- tags: on-device-inference, edge-ai, pytorch, model-export, quantization, microcontrollers, runtime, android, ios, mobile, gpu

## Member repositories
- pytorch/executorch (main) score 95

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:03.701893+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 2026-08-29T18:17:51.422576+00:00, confidence not recorded.
  - readme: https://github.com/pytorch/executorch (fetched 2026-08-28T04:09:03.701893+00:00, sha fec9cbb553e5)
  - homepage: https://executorch.ai (fetched 2026-08-29T08:59:19.118864+00:00, sha 1f6f45fcccbc)
  - registry_pypi: https://pypi.org/pypi/executorch/json (fetched 2026-08-29T08:59:19.128323+00:00, sha 268d86c63d49)
- Data as of 2026-08-30T08:39:29.467469+00:00.