Quantizing RNN Models - 2.0 English

Vitis AI RNN User Guide (UG1563)

Document ID
UG1563
Release Date
2022-01-20
Version
2.0 English

RNN quantizer supports different RNN networks, including RNN, GRU, LSTM, and bi-directional LSTM. It performs fixed-point int16 quantization for model parameters and activations. The compiler generates instructions based on quantized model and hardware architecture.

Note: Currently, RNN compiler only supports standard LSTM, GRU, and OpenIE models, and the generated instructions can only be deployed in the Alveo™ U25 and U50LV cards.