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

4 months ago

ExecuTorch is PyTorch's solution for deploying AI models on-device, supporting smartphones to microcontrollers.
It enables deployment of LLMs, vision, speech, and multimodal models using familiar PyTorch APIs.
Key features include native PyTorch export, a tiny runtime (50KB), and support for 12+ hardware backends.
ExecuTorch uses ahead-of-time (AOT) compilation for model export, compilation, and execution.
It supports various platforms including Android, iOS, Linux, Windows, macOS, and embedded systems.
Includes tools for quantization, memory planning, and developer tools like ETDump profiler.
Provides APIs for multiple languages including Python, C++, Java/Kotlin, and Swift.
Supports advanced features like custom operators, dynamic shapes, and selective build for optimization.
Community contributions are encouraged via GitHub, Discord, and issue reporting.
ExecuTorch is BSD licensed and part of the PyTorch ecosystem.

Hasty Briefsbeta