ExecuTorch 1.0: General Availability Status for Mobile and Embedded...
summary
ExecuTorch 1.0 is GA: a PyTorch-first export-to-edge stack using torch.export and progressive lowering to produce a flatbuffer runtime that runs without a Python or libtorch dependency. It ships native C++, Swift and Kotlin APIs, program/data separation, Windows support, and experimental JS/browser builds. Backend coverage hits 12 targets (ARM Ethos-U, Apple CoreML/MPS, Qualcomm Hexagon, Vulkan GPU, NXP, Samsung, Intel OpenVINO, Cadence, MediaTek) with eight production-ready; deployed across Facebook, WhatsApp, Instagram, Messenger serving billions of inferences. Edge LLM demo (Gemma 3 4B) hits around 8–9 tokens/sec on ARM using xnm pack optimizations.
AI-written summary · about
chapters (16)
- 0:00Welcome to ExecuTorch 1.0
- 1:33On-device AI benefits
- 5:54Design principles and open source
- 7:24ExecuTorch 1.0 timeline
- 7:46General availability announced
- 8:30Success stories and adoption
- 11:01Backend coverage improvements
- 12:47New features in 1.0
- 13:09Demos introduction
- 14:31Gemma 3 on ARM CPU
- 15:56Voxil audio model demo
- 16:39LoRA adapters demo
- 19:05Windows CUDA demo
- 20:52Whisper transcription demo
- 22:02Future directions
- 23:51Closing remarks
related talks
23mBringing ExecuTorch To the Next Frontiers of Edge AI
22mIntroducing ExecuTorch from PyTorch Edge: On-Device AI...
33mExecutorch: Enabling and Accelerating PyTorch Models on Edge Devices
10mExecuTorch on Microcontrollers: Deploying PyTorch To...
20mPyTorch Edge: Developer Journey for Deploying AI Models Onto Edge Devices
13mRunning ExecuTorch Applications With Silicon Accelera...
16mLightning Talk: Efficient Inference at the Edge: Performance You Need at the Lowest...
25mOn-Device LLM Inference on Android With ExecuTorch and Qualcomm QNN