# Accelerate LiteRT Models on Android with KleidiAI and SME2

## In this learning path

- [Introduction](https://learn.arm.com/learning-paths/mobile-graphics-and-gaming/litert-sme/)
- [Explore LiteRT, XNNPACK, KleidiAI, and SME2](https://learn.arm.com/learning-paths/mobile-graphics-and-gaming/litert-sme/1-litert-kleidiai-sme2/)
- [Create LiteRT models](https://learn.arm.com/learning-paths/mobile-graphics-and-gaming/litert-sme/2-build-model/)
- [Build the LiteRT benchmark tool](https://learn.arm.com/learning-paths/mobile-graphics-and-gaming/litert-sme/3-build-tool/)
- [Benchmark the LiteRT model](https://learn.arm.com/learning-paths/mobile-graphics-and-gaming/litert-sme/4-benchmark/)
- [Next Steps](https://learn.arm.com/learning-paths/mobile-graphics-and-gaming/litert-sme/_next-steps/)

## About this Learning Path

|                      |                          |
|----------------------|--------------------------|
| Skill level:         | Advanced                 |
| Reading time:        | 45 min                   |
| Last updated:        | 19 Aug 2026              |

### Author:
Jiaming Guo, Arm

### Arm IP:
[Cortex-A](https://support.arm.com/?tab=compute-ip&Product%20Type=Application%20Processors) [Cortex-X](https://support.arm.com/?tab=compute-ip&Product%20Type=Application%20Processors) [Arm C1](https://support.arm.com/?tab=compute-ip&Product%20Type=Application%20Processors)

### Tags:
[ML](https://learn.arm.com/tag/ml) [Android](https://learn.arm.com/tag/android) [C](https://learn.arm.com/tag/c) [Python](https://learn.arm.com/tag/python) [SME2](https://learn.arm.com/tag/sme2)

### Who is this for?
This is an advanced topic for developers looking to leverage Arm's Scalable Matrix Extension 2 (SME2) instructions to accelerate LiteRT model inference on Android.

### What will you learn?
After completing this Learning Path, you'll be able to perform the following tasks:
- Understand how KleidiAI integrates with LiteRT
- Build the LiteRT benchmark tool and enable XNNPACK and KleidiAI with SME2 support in LiteRT
- Create LiteRT models that can be accelerated by SME2 through KleidiAI
- Use the benchmark tool to evaluate and validate the SME2 acceleration performance of LiteRT models

### Prerequisites
Before starting, you'll need the following prerequisites:
- An Arm64 Linux development machine
- An Android device that supports Arm SME2 architecture features - see this [list of devices with SME2 support](https://learn.arm.com/learning-paths/cross-platform/multiplying-matrices-with-sme2/1-get-started/#devices).

### Summary
You’ll accelerate LiteRT inference on Android with KleidiAI SME2 microkernels through XNNPACK. First, you’ll create models using supported operators and data types, then build a KleidiAI-enabled benchmark and a baseline. After verifying SME2 support, you’ll run both on the same model and compare results, including fallback behavior for unsupported operators.

### Frequently asked questions

**What do I need on the Android device before running benchmarks?**

Copy your LiteRT model file and two `benchmark_model` binaries to the device: one built with KleidiAI and SME2 enabled and one baseline build. Run both against the same model on the same device.

**How do I check whether my Android device supports SME2?**

From an `adb` shell, run `cat /proc/cpuinfo` and look for `sme2` in the `Features` line. If it is present, the CPU supports SME2.

**Which LiteRT operators are accelerated by SME2 through KleidiAI?**

Only the subset of KleidiAI SME2 micro-kernels integrated into XNNPACK are accelerated. Operators outside the supported data types and quantization configurations use XNNPACK’s default implementation.

**What result should I expect when comparing the two benchmark binaries?**

The SME2-enabled binary demonstrates performance gains for models that use supported operators and data types. Both runs should complete successfully so you can compare their reported measurements.

**What should I check if I don’t see an improvement with the SME2-enabled build?**

Verify the device reports SME2 support, and confirm your model uses the supported operator configurations. Also ensure you ran the SME2-enabled binary under the same conditions as the baseline.