How to: Bringing up ExecuTorch on a RISC-V Microcontroller | Corrected Audio
summary
Step‑by‑step work to run ExecuTorch (PyTorch for embedded) on a bare‑metal RISC‑V microcontroller using a Verilator model of the OpenHW CV32E40Pv2 core, adding a new core5 AOT backend and minimal model that adds two 5‑element vectors. Builds include a flatbuffer PTE, runtime hooks in C++ (init/execute/destroy/is_available), and a hex image for Verilator. On the 400 MHz single‑core model with 4 MB microcontroller the full run shows ~9M ticks ≈ 24.82 ms, the actual vector add costs ~16 µs; delegation paths for 8‑bit ops, DMA, double buffering and quantization are integrated though the delegate demo needs one bug fix.
AI-written summary · about
chapters (8)
- 0:00Introduction and goals
- 4:33ExecuTorch overview
- 7:21Environment setup
- 9:31Ahead-of-time processing
- 25:44Runtime implementation
- 41:49Implementing delegation
- 48:13Advanced optimization
- 50:49Summary and conclusion
related talks
10mExecuTorch on Microcontrollers: Deploying PyTorch To...
23mBringing ExecuTorch To the Next Frontiers of Edge AI
13mRunning ExecuTorch Applications With Silicon Accelera...
19mBuilding Tool Chains for RISC-V AI Accelerators
13mAI inference for the smallest processors
22mIntroducing ExecuTorch from PyTorch Edge: On-Device AI...
33mExecutorch: Enabling and Accelerating PyTorch Models on Edge Devices
14mFlexible Deployment of PyTorch Models on MCU-Class...