trionTalk to Trion
DevelopersAnnouncement

ML Drift brings GPU inference to edge devices

Google has open-sourced ML Drift, a GPU compute engine for running AI on devices. It addresses the variety of graphics hardware and runtime APIs that make local inference difficult to ship consistently.

Trion editorialReported 1 min read

One engine across several GPU backends

The ML Drift release supports OpenGL ES, OpenCL, Metal and WebGPU. It is the acceleration engine inside LiteRT and can also be used as a standalone library. The code is released under Apache 2.0.

Google introduced ML Drift, an open source GPU compute engine for AI inference at the edge. It supports graphics and compute backends across platforms, can work with LiteRT or independently, and uses the Apache 2.0 license.
Trion

One inference engine, several GPUs

GPU backends
OpenGL ES, OpenCL, Metal and WebGPU
Packaging
LiteRT engine or standalone library
Primary focus
Mobile and edge; desktop includes previews
License
Apache 2.0
View data
PartPublished support
GPU backendsOpenGL ES, OpenCL, Metal and WebGPU
PackagingLiteRT engine or standalone library
Primary focusMobile and edge; desktop includes previews
LicenseApache 2.0

Google · published 2026-10-08. Source-bound illustration, not a performance benchmark.

Download image

Google describes tensor virtualization, custom operator support and handling for five-dimensional tensors. Its desktop support includes previews, while the primary focus remains mobile and edge devices. Google also says the legacy TensorFlow Lite GPU delegate will no longer receive new feature updates, making migration relevant to existing applications.

Original sources

  1. Google: original announcementdevelopers.googleblog.com

Checked 9 Oct 2026 · A manually curated edition. Availability may change; company performance claims are not Trion test results. Editorial method.