espdl-operator

What problem does it solve?

This Skill streamlines the complex process of developing, optimizing, and validating neural network operators for the ESP-DL framework, ensuring robust and efficient AI model deployment on embedded systems.

Core Features & Use Cases

• End-to-End Operator Lifecycle: Guides through C++ implementation, SIMD optimization, quantization support, and Docker-based testing.
• Automated Validation: Ensures inference results align between quantization tools and on-device runtime.
• Use Case: Implement a new custom operator for a vision model, optimize its performance using SIMD instructions for ESP32-S3, add 8-bit quantization support, and automatically verify its correctness against the esp-ppq toolchain.

Quick Start

Implement the 'HardSigmoid' operator for ESP-DL, including float32 and int8 support, and run comprehensive tests.