xrML

Running quality models on XR devices.

xrML is a community exploring on-device machine learning for extended reality. Our focus is practical ML for augmented, virtual, and mixed reality experiences.

Our focus

• Efficient inference: Exploring models and optimization techniques suited to the compute, memory, and power constraints of XR devices.
• On-device deployment: Investigating how machine learning can run locally within interactive XR applications.
• Practical experimentation: Exploring the tradeoffs between model quality, responsiveness, and resource use on XR hardware.

Supported repositories

We currently support two repositories:

• xrBench: A benchmarking toolkit for measuring native GPU execution time and checking numerical correctness on XR devices.
• xrGPU: Native Vulkan compute kernels and runtime components for machine learning workloads on XR hardware.

Have an XR device?

Visit xrBench to review the device requirements and get started with benchmarking.

Follow our progress

Follow xrML on Hugging Face for updates on machine learning for XR devices.