MLIA
Analyze how your neural network will run on Arm hardware and get actionable advice to improve inference performance. MLIA can also apply model optimizations for supported targets.
Early access
AI Optimization Tooling Early Access
Hardware-aware AI model optimization
Tooling to help you optimize and validate AI models for Arm-based target devices. Optimize models for device compatibility and performance while balancing accuracy goals, then evaluate performance on real devices to speed up application development.
- Optimize for your target device
- Use graph optimization, quantization and model conversion to improve compatibility, speed and memory efficiency.
- Maintain accuracy while improving performance
- Tailor optimization to the target device while minimizing accuracy loss.
- Validate on real devices
- Profile real-device performance and identify bottlenecks with fewer tuning iterations.
Model analysis
Model analysis & optimization Learn moreModel visualization & debug Learn moreModel Explorer
Visualize neural network structures and execution graphs to understand your model before deployment. Inspect architecture, tensor shapes and graph connectivity to help identify issues.
AI performance libraries
AI acceleration Learn moreKleidi AI
Accelerate AI and machine learning workloads on Arm CPUs through optimized kernels integrated into popular frameworks. Get higher performance without changing your application code.
AI acceleration library Learn moreArm Compute Library
A collection of highly optimized machine learning and computer vision functions for Arm CPUs and GPUs. Use it to accelerate common operators and build high-performance ML applications.
Embedded AI acceleration Learn moreCMSIS NN
Optimized neural network kernels for Arm Cortex-M processors, designed to improve inference performance while minimizing memory use on constrained devices.
Mobile and graphics performance tools
Performance profiling Learn moreStreamline
Profile applications running on Arm CPUs and GPUs to understand where time and hardware resources are being spent. Identify performance bottlenecks using software and hardware performance data.
Graphics debugging Learn moreRenderDoc
Capture, inspect and debug graphics applications running on Windows, Linux and Android. Arm's version adds early support for the latest Arm GPU, Vulkan and neural graphics features.