Arm AI Model Optimizer
Arm AI Model Optimizer is a command-line SDK that optimizes AI models for edge deployment. It compresses trained models and restructures their computation graphs. Arm AI Model Optimizer measures accuracy at each checkpoint, then produces a deployment-ready artifact.
How Arm AI Model Optimizer optimizes models
Edge hardware has strict constraints on model size, memory, and compute. Arm AI Model Optimizer automates model optimization so you can improve inference performance without manual quantization or graph transformations.
Given a trained model and a target device, Arm AI Model Optimizer:
- Optionally applies weight-aware quantization at the PyTorch level with Advanced Quantizer
- Fuses and restructures operators for the target hardware with Graph Optimizer
- Reduces numerical precision using calibration data with Graph Quantizer
- Measures accuracy at each checkpoint with Evaluator. For latency and memory profiling, set up your device as described in On-device profiling.
- Exports a
.pteartifact ready for on-device deployment
User guide
The user guide is for model engineers who run the np command-line interface to optimize their models. The guide covers the core concepts, optimization modules, supported models and devices, and the run.yaml configuration. It also includes tutorials and troubleshooting information.
| Goal | Documentation |
|---|---|
| Run your first optimization | Quick start |
| Create or manage an access token | Manage access tokens |
| Understand how it works | Concepts |
| Follow an end-to-end example | Tutorials |
| See available preset models | Supported models |
| Configure the pipeline | run.yaml reference |