SiMa.ai Adds New Features to Palette SDK for ML Developer

SiMa.ai, the software centric, embedded edge machine learning system-on-chip company, announced the latest release of its Palette™ SDK (version 1.3), bringing several enhancements designed to streamline the machine learning (ML) developer experience. This release marks a significant step forward in making ML application development more accessible and efficient for developers worldwide.

Palette is a low-code, command-line environment designed for ML application development on SiMa.ai's Machine Learning System-on-Chip (MLSoC) silicon. It supports the entire application pipeline, from creation to deployment, in minutes using Python scripting. It facilitates auto-partitioning and compilation across the MLA and Quad-core Arm Subsystem with integrated cache. Additionally, developers can integrate C/C++ host applications, libraries, or functions using C/C++ APIs to quickly achieve a cohesive production environment.

With Palette 1.3, SiMa.ai enhances its C++ APIs, improving integration of the MLSoC platform into existing C++ applications. New features include acceleration pipeline status updates and enhanced error reporting, offering increased trace and debug visibility, simplifying the porting of C++ applications using HOST + GPU/Accelerators.

Palette 1.3 also introduces int16 quantization, enhancing precision for hard-to-quantize models. This feature balances precision and computational efficiency, ideal for applications with moderate memory and computational constraints. Developers can start with int8 quantization, assess accuracy using various calibration methods, and switch to int16 quantization if int8 metrics are insufficient, fine-tuning calibration parameters for optimal performance.

Mr. Krishna Rangasayee, CEO and Founder, SiMa.ai commented, “At SiMa.ai, our goal is to support developers at every stage of their ML journey. We continuously add new functionalities, scripts, and models to ensure a seamless and effortless developer experience. With Palette 1.3, we are excited to further empower developers in accelerating their application pipelines at the edge.”

Continuing its commitment to expand and optimize model support, Palette 1.3 now includes MaskRCNN and YOLOv8 4-camera support over Ethernet. This addition joins the more than 350 models already fully compatible with the SiMa.ai MLSoC.