AndesAIRE™ NN SDK
A Comprehensive Set of Software Tools and Runtimes for End-To-End Development and Deployment
AndesAIRE™ NN SDK Overview
AndesAIRE™ (Andes AI Runs Everywhere) NN SDK is a collection of software tools, libraries, runtimes, and sample programs for end-to-end development and deployment of AndesAIRE™ AnDLA™ (Andes Deep Learning Accelerator) and AndesCore™ RISC-V CPU platforms. It includes the following components:
- AndesAIRE™ NNPilot™: a neural network optimization tool suite
- AndesAIRE™ TFLM (TensorFlow Lite for Microcontrollers): an optimized inference framework running on a host
- AnDLA™ driver and runtime
The NNPilot™ can automatically analyze input NN models, apply model pruning and quantization, and generate AnDLA™ executable based on its configuration to perform inference together with the TFLM framework. The NNPilot™ also generates sample host C code to invoke the AnDLA™ driver and AndesAIRE™ NN Library APIs in the bare metal environment.
The product is available in an early-adopter program now.
Features
- Input NN model formats: PyTorch, ONNX, TensorFlow Lite
- Network optimizer: NN model pruning with parameters such as
- Pruning ratio by model/layer
- Automatic, manual, or uniform
- HW-awareness: number of MACs, SRAM size
- Structured pruning
- Deployment optimizer:
- Quantization (INT8): per-layer shift-based, per-layer symmetric, per-channel asymmetric
- Numerical compensation
- Batch normalization fusion
- Backend optimizer:
- Memory footprint optimization, scheduler, OPs fusion, redundant OPs remover
- Performance estimation
- Generated C code template to invoke AnDLA™ driver and NN library APIs
Block Diagram
Related Tools
- AndeSight™ IDE
- AndesAIRE™ NN Library
Product Package
- AndesAIRE™ NN SDK
