The "witin_nn" framework, based on PyTorch, maps neural networks to chip computations and supports operators including Linear, Conv2d, and GruCell. It enables 8-12 bit quantization for inputs/outputs and weights, implementing QAT.
natural-language-processing nat pytorch linear qat nerual-network conv2d computinginmemory convtranspose2d grucell
-
Updated
Aug 15, 2024 - Python