Hands-on TinyML Deployment on RISC-V: Building an Optimized Edge AI Pipeline
summary
Hands-on workshop shows a full pipeline that converts TinyML models to tuned binaries and runs them on RISC-V hardware using an open RISC-V AI software stack and Andes neural network kernel libraries. It demonstrates exporting models from Edge Impulse when available or using Andes model prep tools, optimizing operators and memory layouts, and exercising RISC-V vector and DSP extensions. Example MLPerf Tiny models run live with measured inference improvements and concrete deployment steps and artifacts.
AI-written summary · about
chapters (5)
- 0:00Introduction
- 10:29Setting up the pipeline
- 23:20Deploying on RISC-V
- 30:29Optimizing performance
- 38:42Q&A
related talks
36mTinyML at the Edge: Deploying and Optimizing AI Workloads on Zephyr RTOS
17mExploring Static Code Generation and SIMD-Acceleration for Machine Learning on RISC-V
10mAvoiding Amdahl's Law RISC-V Architecture Exploration for AI & ML Compute
23mEmpower Upstream ML Frameworks on RISC-V
19mAccelerating Edge AI with the Help of RISC-V
14mKeynote: RISC-V Opportunities at the Edge of AI
25mTinyML as-a-Service
10mLightning Talk: Software Development for ML and RISC-V Vector Acceler...