PyTorch's ExecuTorch framework now supports direct deployment of models on Arm Cortex-M microcontrollers, eliminating the need to convert to other framework formats1. The Cortex-M backend uses CMSIS-NN acceleration across the full Cortex-M spectrum from M0 through M85. Benchmarks of four MLPerf Tiny models on the Alif Ensemble E8 show each running in 21.5 ms or less and using under 90 KiB of RAM on a single Cortex-M55 core. The ExecuTorch core runtime occupies under 20 KiB of flash.
ExecuTorch Runs PyTorch Models on Arm Cortex-M MCUs
PyTorch's ExecuTorch now deploys models directly on Arm Cortex-M microcontrollers with CMSIS-NN acceleration, running MLPerf Tiny benchmarks in under…