登录

executorch

pytorch活跃维护

PyTorch 官方的端侧推理框架，把训练好的模型导出并部署到手机、嵌入式和边缘设备上运行，覆盖 CPU、GPU 等多种后端。

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

v1.5.1发布于 2026-09-23 16:58 UTC5,167star +6 / 7 天1,240fork同步于 2026-11-10 05:36 UTC
端侧推理模型部署边缘计算PyTorch嵌入式
对比 GitHub 官网deep-learningembeddedgpumachine-learningmobile

怎么上手

由 README 提炼，执行前请确认
原文 
pip install executorch
硬件要求按 README 说明

README 未给出具体显存或内存门槛；Linux 上 pip 安装的是 CUDA 构建，macOS 和 Windows 上是 CPU 构建，CUDA 版 wheel 仅发布 Linux 版本；嵌入式侧覆盖 Cortex-M、Ethos-U、NXP NPU、Cadence DSP 等目标。

纯 CPU 可运行需要 CUDA
下载：源码 zip·GitHub Releases（安装包）本站不镜像安装包

项目说明

由 README 自动翻译提炼
没人工点评？写下你的经验

PyTorch 官方的端侧推理栈，把 PyTorch 模型导出成 .pte 后跑在手机、嵌入式设备和微控制器上

PyTorch 官方的端侧推理框架，把训练好的模型导出并部署到手机、嵌入式和边缘设备上运行，覆盖 CPU、GPU 等多种后端。

核心能力

  • 用 torch.export 捕获模型并导出为 .pte 文件
  • 按硬件划分计算图，把子图交给 CPU、GPU、NPU 或 DSP 后端
  • 支持 selective build，只链接部署所需的算子内核
  • 提供 ETDump、ETRecord 和数值调试工具做性能与精度排查
  • 支持 C++、Python、Swift、Kotlin、JavaScript 多语言运行时 API

容易踩的坑

需要 Python 3.11–3.15 环境；nightly 版 wheel 不声明 torch 依赖，必须显式一起安装 torch，否则可能被替换成 CPU 版；Ethos-U 等后端导出要另装可选依赖如 executorch[ethos_u]，嵌入式工具链需单独安装；.pte 与所选后端绑定，换硬件要重新导出而非只换 partitioner。

内容由模型根据 README 生成，可能不完整或不准确，以官方文档为准。看原文 README

最近发版

5 个版本v1.5.1
全部 release
  1. v1.5.1最新正式版2026-09-23 · 2026-09-23 16:58 UTC

    v1.5.1 Changes

    • Updated CMSIS-NN to v8.0.0 and aligned Arm runner dependencies.
    • Added the public get_thread_count() API.
    • Expanded Arm support for symbolic shapes and dynamic convolution and pooling.
    • Folded redundant Q/DQ pairs in the shared EXIR to_edge lowering path.

    Highlights

    • Expanded the Python wheels into a C++ SDK, with linkable libraries for the CUDA, Core ML, MLX, OpenVINO, and Qualcomm delegates and packaged TorchAO kernels.
    • Expanded LLM serving with multi-method export, batched request scheduling, bounded cancellation, off-graph KV-cache layouts, and new Qwen3.5 MoE, Muse Glimmer, Supertonic, and Voxtral workflows.
    • Added ROCm support to the AOTI CUDA backend, including packed W4 execution and end-to-end Voxtral Realtime coverage.
    • Expanded Arm and Cortex-M support with explicit-layout lowering and kernels, additional Ethos-U targets and demos, broader VGF profiling, and more model and operator coverage.
    • Expanded Qualcomm LLM quantization-aware distillation, additional operators, Windows guidance, and stronger validation and error handling.
    • Improved Vulkan quantized execution, dispatch safety, external-memory integration, and operator coverage.

    Runtime and Core

    • Added multi-method model support to the export pipeline and per-backend load options.
    • Added capacity-aware memory planning over target memory maps and support for shared allocations with offsets.
    • Added schema-version checks for PTE and PTD files and clearer failures for invalid device-planned copies and missing delegate data.
    • Added shared pre-decomposition and graph-transform hooks, including common view/permute propagation and reinplace passes.
    • Improved constant propagation, symbolic-shape handling, and device placement for memory-planned buffers.

    Packaging and Build

    • Added a C++ SDK to the Python wheel and split the runtime from the Python bindings extension.
    • Added CUDA wheel publishing on Pytorch.org
    • Improved delegate/runtime bundling in the published wheels.
    • Added packaged TorchAO and quantized-kernel libraries.
    • Added Windows ARM64 build coverage.
    • Reduced wheel size and improved portability by stripping local macOS symbols, removing unreachable vendored sources, and dropping unusable Linux MKL search paths.
    • Lowered the supported CMake floor so CMake 3.26 through 3.28 builds work again.

    Backends

    Arm

    • Added opt-in Cortex-M explicit-layout lowering with convolution and pooling kernels.
    • Added Arm export-recipe support, a pre-decomposition partitioner pipeline, and broader symbolic-shape and dim-order handling.
    • Added Ethos-U65 support, Ethos-U85 KV-cache export, and SmolLM2, TinyStories-42M, Silero VAD, MobileSAM, and Stable Diffusion workflows.
    • Expanded VGF profiling and environment preflight validation tools, neural statistics, automatic operator-support generation, support for int8 KV cache, FP16, BF16, MXFP8, and quantized model coverage.
    • Improved Cortex-M memory planning, CMSIS-Pack coverage, in-place activation fusion, and arithmetic, pooling, and activation lowering.

    Qualcomm

    • Added LLM quantization-aware distillation and Gemma 4 text-model enablement.
    • Added support for pdist, sort, reflection_pad3d, as_strided, scatter_reduce, scatter_add, and Hadamard-transform flows.
    • Added QNN-on-Windows guidance and improved SDK setup, import behavior, delegate bounds checks, graph-I/O validation, and quantized metadata preservation.

    CUDA and ROCm

    • Added opt-in ROCm support to the AOTI backend, packed W4 execution, AMD pointwise examples, and Voxtral Realtime coverage on gfx1100 and gfx950.
    • Added shared-memory targeting for cross-architecture CUDA inference and INT4, INT5, and INT6 short-query kernels.
    • Improved CUDA delegate integr
  2. v1.5.02026-09-16 · 2026-09-16 16:08 UTC
    查看更新说明

    Highlights

    • Expanded the Python wheels into a C++ SDK, with linkable libraries for the CUDA, Core ML, MLX, OpenVINO, and Qualcomm delegates and packaged TorchAO kernels.
    • Expanded LLM serving with multi-method export, batched request scheduling, bounded cancellation, off-graph KV-cache layouts, and new Qwen3.5 MoE, Muse Glimmer, Supertonic, and Voxtral workflows.
    • Added ROCm support to the AOTI CUDA backend, including packed W4 execution and end-to-end Voxtral Realtime coverage.
    • Expanded Arm and Cortex-M support with explicit-layout lowering and kernels, additional Ethos-U targets and demos, broader VGF profiling, and more model and operator coverage.
    • Expanded Qualcomm LLM quantization-aware distillation, additional operators, Windows guidance, and stronger validation and error handling.
    • Improved Vulkan quantized execution, dispatch safety, external-memory integration, and operator coverage.

    Runtime and Core

    • Added multi-method model support to the export pipeline and per-backend load options.
    • Added capacity-aware memory planning over target memory maps and support for shared allocations with offsets.
    • Added schema-version checks for PTE and PTD files and clearer failures for invalid device-planned copies and missing delegate data.
    • Added shared pre-decomposition and graph-transform hooks, including common view/permute propagation and reinplace passes.
    • Improved constant propagation, symbolic-shape handling, and device placement for memory-planned buffers.

    Packaging and Build

    • Added a C++ SDK to the Python wheel and split the runtime from the Python bindings extension.
    • Added CUDA wheel publishing on Pytorch.org
    • Improved delegate/runtime bundling in the published wheels.
    • Added packaged TorchAO and quantized-kernel libraries.
    • Added Windows ARM64 build coverage.
    • Reduced wheel size and improved portability by stripping local macOS symbols, removing unreachable vendored sources, and dropping unusable Linux MKL search paths.
    • Lowered the supported CMake floor so CMake 3.26 through 3.28 builds work again.

    Backends

    Arm

    • Added opt-in Cortex-M explicit-layout lowering with convolution and pooling kernels.
    • Added Arm export-recipe support, a pre-decomposition partitioner pipeline, and broader symbolic-shape and dim-order handling.
    • Added Ethos-U65 support, Ethos-U85 KV-cache export, and SmolLM2, TinyStories-42M, Silero VAD, MobileSAM, and Stable Diffusion workflows.
    • Expanded VGF profiling and environment preflight validation tools, neural statistics, automatic operator-support generation, support for int8 KV cache, FP16, BF16, MXFP8, and quantized model coverage.
    • Improved Cortex-M memory planning, CMSIS-Pack coverage, in-place activation fusion, and arithmetic, pooling, and activation lowering.

    Qualcomm

    • Added LLM quantization-aware distillation and Gemma 4 text-model enablement.
    • Added support for pdist, sort, reflection_pad3d, as_strided, scatter_reduce, scatter_add, and Hadamard-transform flows.
    • Added QNN-on-Windows guidance and improved SDK setup, import behavior, delegate bounds checks, graph-I/O validation, and quantized metadata preservation.

    CUDA and ROCm

    • Added opt-in ROCm support to the AOTI backend, packed W4 execution, AMD pointwise examples, and Voxtral Realtime coverage on gfx1100 and gfx950.
    • Added shared-memory targeting for cross-architecture CUDA inference and INT4, INT5, and INT6 short-query kernels.
    • Improved CUDA delegate integration with per-thread streams, a warm memory pool, cross-method weight sharing, and clearer missing-weight errors.

    Core ML and MLX

    • Added linkable Core ML and MLX libraries to Apple frameworks and SwiftPM, plus ETDump profiling support.
    • Added MLX off-graph KV-cache flat, ring
  3. v1.4.12026-08-14 · 2026-08-14 21:07 UTC
    查看更新说明

    1.4.1 Changes

    Fixed an issue with incorrect baked in paths within the wheels. Fixed an issue with XNNPack's PreLU in quantized graphs.

    1.4 Highlights

    • Expanded Cortex-M support across additional CPU variants, model tests, scratch-buffer planning, CMSIS-NN integration, and CMSIS Pack builds.
    • Added broad Arm backend coverage for TOSA dialect lowering, dynamic shapes, FP8/MXFP/BFloat16 paths, VGF profiling, and LLM export workflows.
    • Added Vulkan support for Linux and Windows desktop GPUs, opt-in wheel builds, automatic build enablement, and a functional backend availability check.
    • Expanded Qualcomm QNN support with 2-bit 16a2w quantization, faster VLM/LLM PTQ flows, multi-batch quantization and evaluation, LPAI tooling, profiling, and additional operators.
    • NXP Backend finished MLIR based Neutron Converter flow rollout - all operators using the MLIR flow, Neutron backend now support 48 ATen operators.
    • Continued Android API migration to Kotlin and added Android model E2E coverage and Kotlin examples.

    Runtime and Core

    • Removed the legacy _use_edge_ops toggles from the runtime/export path.
    • Added explicit dim-order legalization for channels-last portable fallbacks.
    • Hardened tensor-layout validation and segment loading against null metadata.
    • Improved symbolic input metadata preservation during exported-program pass replay.
    • Added direct bundled-program serialization improvements for embedded flows.

    Backends

    Arm and Cortex-M

    • Added builds for more Cortex-M variants on Corstone-300, including M0+.
    • Added AoT scratch-buffer planning and broader MLPerf Tiny model coverage.
    • Expanded TOSA dialect lowering for activation, binary, conversion, data layout, FFT, reduction, shape, and unary operators.
    • Added or improved VGF support for profiling, shader segments, BFloat16, MXFP, grid sampling, and persistent IO mappings.
    • Ethos-U and VGF dependencies can now be installed with ExecuTorch using optional packages “pip install executorch[ethos_u,vgf]”
    • Added SmolLM2 and Qwen3-VL export and evaluation coverage.
    • Added CMSIS Pack infrastructure and an ExecuTorch 1.4 public API manifest.

    Qualcomm

    • Added additional QNN operators, including addmm, select_scatter, variance, diagonal, fill, cdist, LayerNorm variants, and hyperbolic functions.
    • Added 2-bit weight quantization support and multi-batch LLM quantization.
    • Decoupled calibration, quantization, compilation, and evaluation stages for faster LLM and VLM workflows.
    • Added LPAI partitioning, tensor dumping, and runtime heap profiling.

    NXP

    • Updated the Neutron software stack to 3.1.3.
    • Added backend profiling and MLPerf Tiny profiling tests.
    • Expanded Neutron operator coverage and removed several broadcasting restrictions.
    • Fixed partition-output handling, format inference, BatchNorm fusion, and multiprocessing conversion issues.

    CoreML and MLX

    • Added a Core ML compute-plan reporting tool for ANE, GPU, and CPU dispatch.
    • Added MLX handlers for hardtanh, bitwise_or, and leaky_relu.
    • Improved MLX submodule build isolation.

    Vulkan

    • Added Vulkan desktop support for Linux and Windows and opt-in wheel builds.

    XNNPack

    • Added quantized same-padding convolution support in XNNPACK, including fixes for even Conv1d kernels and ViT concatenation quantization parameters.
    • Expanded RISC-V XNNPACK coverage across multiple RVV QEMU configurations.

    Operators and Kernels

    • Added BFloat16 and FP16 activation support to CPU SDPA.
    • Added BFloat16-output quantized embeddings on CPU.
    • Improved numerical accuracy by accumulating Half and BFloat16 softmax, log-softmax, mean, and sum operations in FP32.
    • Fixed portable anti-aliased bilinear upsampling correctness.

    LLM, Mobile, and Examples

    • Added grammar and grammar_type
  4. v1.4.0v1.42026-08-07 · 2026-08-07 19:06 UTC
    查看更新说明

    Highlights

    • Expanded Cortex-M support across additional CPU variants, model tests, scratch-buffer planning, CMSIS-NN integration, and CMSIS Pack builds.
    • Added broad Arm backend coverage for TOSA dialect lowering, dynamic shapes, FP8/MXFP/BFloat16 paths, VGF profiling, and LLM export workflows.
    • Added Vulkan support for Linux and Windows desktop GPUs, opt-in wheel builds, automatic build enablement, and a functional backend availability check.
    • Expanded Qualcomm QNN support with 2-bit 16a2w quantization, faster VLM/LLM PTQ flows, multi-batch quantization and evaluation, LPAI tooling, profiling, and additional operators.
    • NXP Backend finished MLIR based Neutron Converter flow rollout - all operators using the MLIR flow, Neutron backend now support 48 ATen operators.
    • Continued Android API migration to Kotlin and added Android model E2E coverage and Kotlin examples.

    Runtime and Core

    • Removed the legacy _use_edge_ops toggles from the runtime/export path.
    • Added explicit dim-order legalization for channels-last portable fallbacks.
    • Hardened tensor-layout validation and segment loading against null metadata.
    • Improved symbolic input metadata preservation during exported-program pass replay.
    • Added direct bundled-program serialization improvements for embedded flows.

    Backends

    Arm and Cortex-M

    • Added builds for more Cortex-M variants on Corstone-300, including M0+.
    • Added AoT scratch-buffer planning and broader MLPerf Tiny model coverage.
    • Expanded TOSA dialect lowering for activation, binary, conversion, data layout, FFT, reduction, shape, and unary operators.
    • Added or improved VGF support for profiling, shader segments, BFloat16, MXFP, grid sampling, and persistent IO mappings.
    • Ethos-U and VGF dependencies can now be installed with ExecuTorch using optional packages “pip install executorch[ethos_u,vgf]”
    • Added SmolLM2 and Qwen3-VL export and evaluation coverage.
    • Added CMSIS Pack infrastructure and an ExecuTorch 1.4 public API manifest.

    Qualcomm

    • Added additional QNN operators, including addmm, select_scatter, variance, diagonal, fill, cdist, LayerNorm variants, and hyperbolic functions.
    • Added 2-bit weight quantization support and multi-batch LLM quantization.
    • Decoupled calibration, quantization, compilation, and evaluation stages for faster LLM and VLM workflows.
    • Added LPAI partitioning, tensor dumping, and runtime heap profiling.

    NXP

    • Updated the Neutron software stack to 3.1.3.
    • Added backend profiling and MLPerf Tiny profiling tests.
    • Expanded Neutron operator coverage and removed several broadcasting restrictions.
    • Fixed partition-output handling, format inference, BatchNorm fusion, and multiprocessing conversion issues.

    CoreML and MLX

    • Added a Core ML compute-plan reporting tool for ANE, GPU, and CPU dispatch.
    • Added MLX handlers for hardtanh, bitwise_or, and leaky_relu.
    • Improved MLX submodule build isolation.

    Vulkan

    • Added Vulkan desktop support for Linux and Windows and opt-in wheel builds.

    XNNPack

    • Added quantized same-padding convolution support in XNNPACK, including fixes for even Conv1d kernels and ViT concatenation quantization parameters.
    • Expanded RISC-V XNNPACK coverage across multiple RVV QEMU configurations.

    Operators and Kernels

    • Added BFloat16 and FP16 activation support to CPU SDPA.
    • Added BFloat16-output quantized embeddings on CPU.
    • Improved numerical accuracy by accumulating Half and BFloat16 softmax, log-softmax, mean, and sum operations in FP32.
    • Fixed portable anti-aliased bilinear upsampling correctness.

    LLM, Mobile, and Examples

    • Added grammar and grammar_type fields to GenerationConfig.
    • Added an unoptimized ExecuTorch runner example for Espressif ESP32.
    • Migrated Android Module, TrainingModule, SGD,
  5. v1.3.12026-05-29 · 2026-05-29 17:34 UTC
    查看更新说明

    ExecuTorch v1.3.1 Release Notes

    1.3.1 is the first broadly published 1.3 patch release, following the Maven-only 1.3.0 release.

    Highlights

    ExecuTorch v1.3.1 expands model and backend coverage across embedded, mobile, and GPU targets. This release adds major Arm, Cortex-M, VGF, NXP, Qualcomm, CUDA, Metal, MLX, Vulkan, and XNNPACK improvements, and continues broadening LLM and multimodal model support.

    • Arm backend (Ethos-U / TOSA / VGF / Cortex-M) - Major expansion in TOSA dialect lowering, quantization support, VGF infrastructure, Cortex-M test coverage, Ethos-U compatibility, and CI reliability.
    • New and expanded model support - Qwen3.5 MoE, Gemma 4 31B, LFM2.5, Voxtral Realtime/TTS, Llama4 export support, and additional Android LLM runner documentation.
    • CUDA backend - Qwen3.5 MoE export/runtime work, fused MoE and INT4 Triton kernels, CUDA graph capture/replay, GPU-side sampling, and memory/stat reporting.
    • Metal and MLX backends - Qwen3.5 MoE Metal path, Gemma 4 31B MLX support, MLX delegate progress, integer op support, top-k fallback, gated delta rule kernels, and hardened runtime code signing.
    • Vulkan backend - Cooperative matrix dispatch for linear/matmul, safer memory handling, integer overflow and heap-buffer fixes, improved partitioning, and VGF compatibility work.
    • Qualcomm AI Engine Direct (QNN) - Expanded ATen op support, LPAI and Direct Mode work, custom op and quantization annotation APIs, debugging improvements, performance dumping, and heap profiling.
    • NXP backend - Continued new Neutron flow rollout, eIQ Neutron SDK 3.1.1, stricter graph verification, QAT coverage, and additional operator support.
    • Runtime and packaging - Shared library install support, Android lifecycle/error handling fixes, and improved wheel/build reliability.

    New Models and Model Enablement

    • Qwen3.5 MoE export and runner support across CUDA and Metal, including fused MoE kernels, INT4 paths, CUDA graph support, and runtime stats.
    • Gemma 4 31B export and runner support with CUDA and MLX backend support.
    • LFM2.5 export and runner support for MLX.
    • Voxtral Realtime and Voxtral TTS improvements, including streaming fixes, CUDA backend support, and documentation updates.
    • Android LLM runner documentation and HuggingFace-oriented usage docs.

    Core Components

    EXIR / Export

    • Safer ETRecord/export deserialization.
    • FP8 placeholder support in ExecuTorch serialization.
    • Fixes for torch.split alias annotations and partition grouping.
    • Migration cleanup away from deprecated capture config paths.

    Runtime

    • Backend and runtime option handling improvements.
    • Better contextual error reporting for debuggability.

    Kernels and Operators

    • Portable adaptive average pool 2D kernel.
    • fp32 accumulation for Half/BFloat16 grid sampler bilinear paths.
    • Fused quantized add, mul, relu, hardswish, and bmm kernels.
    • Quantized max pool and quantized conv1d fixes.
    • Additional quantization and observer fixes across portable and backend kernels.

    Backend Delegates

    Arm / Ethos-U / TOSA / VGF / Cortex-M

    • Expanded TOSA op coverage including slice, max pool, avg pool, adaptive avg pool, shape ops, resize, LSTM decomposition, and additional dtype/layout handling.
    • FP16/BF16 operator coverage is now considered largely complete; most support landed incrementally over prior releases and the 1.3 cycle adds test stabilization for FP16 MobileNetV3 and VGF bilinear paths.
    • Refactored dim-order and permute handling. Removed the legacy tosa_dim_order path, added permute-removal for TOSA ops, and improved permute fusion over elementwise ops. This reduced unnecessary transposes in lowered graphs.
    • Improved quantization support, including composable quantizer work, constant folding fixes, TOSA/VGF linear quantization modes, and Cortex-M MVE/Helium int16 quantize/dequantize support.
    • VGF build/test infrastructure and documentation.

用户点评

登录用户的真实使用记录，可按版本和硬件筛选
共 0 条

    还没有人点评。第一个说出真实体验的人最有价值。

    README 原文

    未翻译
    在 GitHub 打开

    ExecuTorch is PyTorch's open source stack for running AI locally on phones, wearables, laptops, browsers, embedded systems, and microcontrollers. Start with a PyTorch model, capture it with torch.export, optimize it for target hardware, and run it through C++, Python, Swift/Objective-C, Kotlin/Java, or JavaScript APIs.

    ExecuTorch powers on-device experiences across Instagram, WhatsApp, Facebook, and Messenger, serving billions of people. It also runs AI features on Meta Quest and Ray-Ban Meta devices. See where ExecuTorch is shipping.

    [!IMPORTANT] Release channels: use the latest release with the stable documentation. This README tracks main; features labeled Main / nightly may change before release. Use the main documentation with a source checkout or nightly package.

    Built for current edge workloads

    WorkloadRepresentative capabilitiesStart here
    Local LLM and agent building blocksQuantized text models, long context, tool calling, speculative decoding, multi-session execution, and an experimental OpenAI-compatible local serverMuse Glimmer · LLM guide · LLM server
    VoiceStreaming and offline speech recognition, speech synthesis, voice activity detection, and speaker diarizationVoxtral Realtime · Parakeet · Sortformer diarization · Voxtral TTS · Nemotron 3 Diarization
    MultimodalText, image, and audio runners; mobile vision-language modelsGemma 4 · Multimodal runner
    Computer visionImage classification, object detection, semantic segmentation, and promptable segmentationMobileNet V2 · YOLO26 · DeepLabV3 · EfficientSAM
    Embedded AICortex-M CPU kernels, Ethos-U and NXP NPUs, Cadence DSPs, Zephyr, Arduino, and Raspberry Pi Pico workflowsEmbedded guide · Cortex-M · Arduino

    These are starting points, not a compatibility list. A model does not need to appear here to run with ExecuTorch: use the standard export guide or adapt the closest model example. Deployment depends on torch.export capture plus operator, runtime, and selected-backend coverage; validate numerical accuracy, memory use, and performance on the target. For a new language-model architecture, see the custom LLM guide.

    Why ExecuTorch

    • PyTorch-native workflow: work directly from torch.export, with PyTorch program metadata and source mappings available to export and debugging tools.
    • Partitioned hardware acceleration: delegate supported graph regions to CPU, GPU, NPU, or DSP backends while retaining portable CPU kernels for fallback.
    • One source model, explicit targets: reuse the PyTorch model and export flow, while producing a backend-specific .pte for each target that needs hardware specialization.
    • A runtime you can right-size: link only the operators, kernels, and delegates a deployment needs; selective build keeps only the required operator kernels.
    • Inspect, profile, and extend: use ETDump, ETRecord, and numeric debugging, or add custom operators and backends.
    • Versioned deployment contract: a .pte created with stable APIs is guaranteed to load and execute for at least one following non-patch runtime release; see the runtime compatibility and API lifecycle policies.

    Install

    Install the latest stable Python package in a Python 3.11–3.15 environment:

    pip install executorch

    On Linux this is the CUDA build, matching the torch package on PyPI. On macOS and Windows it is the CPU build. For the CPU-only Linux build or another CUDA version, see Getting Started.

    Install a nightly built from main to use the newest features:

    pip install --upgrade --pre executorch torch --extra-index-url https://download.pytorch.org/whl/nightly/cpu

    torch is explicit because nightly ExecuTorch wheels do not declare it as a dependency. This command installs a CPU-only PyTorch nightly. On Linux with an NVIDIA GPU, select the nightly command matching your CUDA version in the PyTorch installation selector and use its nightly/cu* index instead; otherwise, pip can replace a CUDA-enabled PyTorch installation with the CPU build. ExecuTorch CUDA wheels are published for Linux only.

    Backend export tools can require optional dependencies. For example, use pip install 'executorch[ethos_u]' for Ethos-U AOT export. Embedded toolchains, simulators, and target runtimes are installed separately.

    For platform-specific setup (Android, iOS, embedded systems), see the Quick Start documentation for additional information.

    The prebuilt Python wheel is published for Linux x86-64, Linux AArch64, macOS arm64, and Windows x86-64. Build from source for other hosts or custom configurations. Native integration is also available through:

    Five-minute export and run

    This complete example, from the quick-start pathway, exports a small model and lowers it to XNNPACK:

    import torch
from executorch.exir import to_edge_transform_and_lower
from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
from executorch.runtime import Runtime

class Add(torch.nn.Module):
    def forward(self, x, y):
        return x + y

model = Add().eval()
sample_inputs = (torch.ones(1), torch.ones(1))

et_program = to_edge_transform_and_lower(
    torch.export.export(model, sample_inputs),
    partitioner=[XnnpackPartitioner()]
).to_executorch()

with open("add.pte", "wb") as f:
    f.write(et_program.buffer)

runtime = Runtime.get()
runtime_program = runtime.load_program("add.pte")
method = runtime_program.load_method("forward")
output = method.execute(sample_inputs)[0]

torch.testing.assert_close(output, model(*sample_inputs))
print("Output:", output)

    Expected output: Output: tensor([2.]). This verifies export, XNNPACK lowering, serialization, loading, and execution on the host; measure performance again on the target device.

    The resulting add.pte is specialized for the selected backend. Targeting Core ML, Qualcomm, or another accelerator requires that backend's dependencies, configuration, and a separate export, not a blind partitioner substitution. Continue with the Python runtime, C++ Module API, Android, or iOS.

    Choose an export path

    Starting pointRecommended pathScope
    PyTorch nn.ModuleStandard export and loweringFull backend choice; some models need decompositions or custom operators
    Hugging Face PreTrainedModelTransformers ExecuTorch exporterExperimental programmatic XNNPACK/CUDA export; generation components require application orchestration
    Optimized text or multimodal generationexport_llm or Optimum ExecuTorchTested recipes, quantization, tokenizers, and runner-specific metadata

    Already have a compatible .pte? Browse the ExecuTorch Community, Arm AI model catalog filtered to ExecuTorch, or, where available, the linked model pages above. Match the model configuration, precision, backend, and runtime; a .pte built for one hardware delegate is not a universal model file.

    Runtime and LLM APIs

    Use C++, Python, Java/Kotlin, Swift/Objective-C, or JavaScript/WebAssembly for general .pte execution. Higher-level text and multimodal runners are available for C++, Python, Android, and Apple platforms.

    Compare the runtime and LLM APIs for language-specific entry points and API maturity.

    Platforms and hardware backends

    Choose a backend based on target hardware, operator coverage, and deployment constraints. The linked guides document setup, supported hardware, and known limitations. These are representative paths; the backend documentation is authoritative. See also the Desktop guide.

    TargetBackends and integrations
    AndroidXNNPACK CPU; Vulkan GPU; Qualcomm, MediaTek, Arm VGF, and Samsung Exynos accelerators
    iOS / iPadOSXNNPACK CPU; Core ML; MLX on physical devices (experimental)
    macOSXNNPACK; Core ML; experimental MLX, Metal/AOTInductor, and WebGPU paths
    LinuxXNNPACK; OpenVINO; experimental CUDA/AOTInductor, Vulkan, and WebGPU desktop paths
    WindowsXNNPACK; OpenVINO; experimental CUDA/AOTInductor and Vulkan desktop paths
    Browser / WebAssemblyPortable WebAssembly runtime and WebGPU (both experimental)
    Embedded / MCUArm Cortex-M with CMSIS-NN (beta); Arm Ethos-U; NXP eIQ Neutron; Cadence DSP; Zephyr and Arduino integrations

    Documentation

    Community and contributing

    Citing ExecuTorch

    Read the MLSys 2026 paper (PDF).

    If you use ExecuTorch in research, please cite:

    @article{executorch2026,
    title={{ExecuTorch} - A Unified {PyTorch} Solution to Run {AI} Models On-Device},
    author={Nachin, Mergen and Desai, Digant and Jia, Sicheng Stephen and Lai, Chen and Liu, Mengwei and Szwejbka, Jacob and Alvarez, Raziel and Ascani, RJ and Bort, Dave and Candales, Manuel and
  others},
    journal={arXiv preprint arXiv:2605.08195},
    url={https://arxiv.org/abs/2605.08195},
    year={2026}
  }

    License

    ExecuTorch is BSD licensed. See LICENSE.

    相关项目

    同分类里 star 最高的项目
    对比全部 →
    mediapipe
    google-ai-edge·C++

    跨平台机器学习框架，提供可定制的视觉、音频等实时流媒体处理方案，方便在移动端和桌面端搭建低延迟推理管线。

    端侧推理+4端侧部署
    37624+356.2kv1.2.22026-11-10 04:08 UTC
    ncnn
    Tencent·C++

    面向手机端的高性能神经网络推理框架，针对 ARM 平台深度优化，让 Caffe、ONNX 等模型能在 Android/iOS 上低延迟运行。

    推理框架+4端侧部署
    24032+24.5k202610202026-11-10 04:45 UTC
    mlc-llm
    mlc-ai·Python

    通用大模型部署引擎，借助机器学习编译技术把 LLM 高效跑在各种硬件上，从手机到服务器都能用。

    推理引擎+4端侧部署
    23318+82.2kv0.1.dev02026-10-30 09:07 UTC
    MNN
    alibaba·C++

    阿里开源的轻量级端侧推理引擎，主打手机和嵌入式设备上的高性能模型推理，能跑大模型和常规深度学习网络。

    推理引擎+4端侧部署
    16466+432.5k3.6.22026-11-10 05:03 UTC