CMSIS-NN  
CMSIS NN Software Library
Data Fields
cmsis_nn_lstm_params Struct Reference

Data Fields

int32_t time_major
 
cmsis_nn_scaling input_to_input_scaling
 
cmsis_nn_scaling input_to_forget_scaling
 
cmsis_nn_scaling input_to_cell_scaling
 
cmsis_nn_scaling input_to_output_scaling
 
cmsis_nn_scaling recurrent_to_input_scaling
 
cmsis_nn_scaling recurrent_to_forget_scaling
 
cmsis_nn_scaling recurrent_to_cell_scaling
 
cmsis_nn_scaling recurrent_to_output_scaling
 
cmsis_nn_scaling cell_to_input_scaling
 
cmsis_nn_scaling cell_to_forget_scaling
 
cmsis_nn_scaling cell_to_output_scaling
 
cmsis_nn_scaling projection_scaling
 
cmsis_nn_scaling hidden_scaling
 
cmsis_nn_scaling layer_norm_input_scaling
 
cmsis_nn_scaling layer_norm_forget_scaling
 
cmsis_nn_scaling layer_norm_cell_scaling
 
cmsis_nn_scaling layer_norm_output_scaling
 
int32_t cell_state_shift
 
int32_t hidden_offset
 
int32_t output_state_offset
 
cmsis_nn_lstm_clip_params clip
 
cmsis_nn_lstm_guard_params guard
 
cmsis_nn_layer_norm layer_norm
 
const int32_t * i2i_effective_bias
 
const int32_t * i2f_effective_bias
 
const int32_t * i2c_effective_bias
 
const int32_t * i2o_effective_bias
 
const int32_t * r2i_effective_bias
 
const int32_t * r2f_effective_bias
 
const int32_t * r2c_effective_bias
 
const int32_t * r2o_effective_bias
 
const int32_t * projection_effective_bias
 
const int32_t * input_gate_bias
 
const int32_t * forget_gate_bias
 
const int32_t * cell_gate_bias
 
const int32_t * output_gate_bias
 
cmsis_nn_activation activation
 

Description

Parameters for integer LSTM, as defined in TFLM

Field Documentation

◆ activation

cmsis_nn_activation activation

◆ cell_gate_bias

const int32_t* cell_gate_bias

◆ cell_state_shift

int32_t cell_state_shift

◆ cell_to_forget_scaling

cmsis_nn_scaling cell_to_forget_scaling

◆ cell_to_input_scaling

cmsis_nn_scaling cell_to_input_scaling

◆ cell_to_output_scaling

cmsis_nn_scaling cell_to_output_scaling

◆ clip

cmsis_nn_lstm_clip_params clip

◆ forget_gate_bias

const int32_t* forget_gate_bias

◆ guard

cmsis_nn_lstm_guard_params guard

◆ hidden_offset

int32_t hidden_offset

◆ hidden_scaling

cmsis_nn_scaling hidden_scaling

◆ i2c_effective_bias

const int32_t* i2c_effective_bias

input to cell effective bias

◆ i2f_effective_bias

const int32_t* i2f_effective_bias

input to forget gate effective bias

◆ i2i_effective_bias

const int32_t* i2i_effective_bias

input to input effective bias

◆ i2o_effective_bias

const int32_t* i2o_effective_bias

input to output effective bias

◆ input_gate_bias

const int32_t* input_gate_bias

◆ input_to_cell_scaling

cmsis_nn_scaling input_to_cell_scaling

◆ input_to_forget_scaling

cmsis_nn_scaling input_to_forget_scaling

◆ input_to_input_scaling

cmsis_nn_scaling input_to_input_scaling

◆ input_to_output_scaling

cmsis_nn_scaling input_to_output_scaling

◆ layer_norm

cmsis_nn_layer_norm layer_norm

◆ layer_norm_cell_scaling

cmsis_nn_scaling layer_norm_cell_scaling

layer normalization for cell

◆ layer_norm_forget_scaling

cmsis_nn_scaling layer_norm_forget_scaling

layer normalization for forget gate

◆ layer_norm_input_scaling

cmsis_nn_scaling layer_norm_input_scaling

layer normalization for input layer

◆ layer_norm_output_scaling

cmsis_nn_scaling layer_norm_output_scaling

layer normalization for outpus layer

◆ output_gate_bias

const int32_t* output_gate_bias

◆ output_state_offset

int32_t output_state_offset

◆ projection_effective_bias

const int32_t* projection_effective_bias

◆ projection_scaling

cmsis_nn_scaling projection_scaling

◆ r2c_effective_bias

const int32_t* r2c_effective_bias

recurrent gate to cell effective bias

◆ r2f_effective_bias

const int32_t* r2f_effective_bias

recurrent gate to forget gate effective bias

◆ r2i_effective_bias

const int32_t* r2i_effective_bias

recurrent gate to input effective bias

◆ r2o_effective_bias

const int32_t* r2o_effective_bias

recurrent gate to output effective bias

◆ recurrent_to_cell_scaling

cmsis_nn_scaling recurrent_to_cell_scaling

◆ recurrent_to_forget_scaling

cmsis_nn_scaling recurrent_to_forget_scaling

◆ recurrent_to_input_scaling

cmsis_nn_scaling recurrent_to_input_scaling

◆ recurrent_to_output_scaling

cmsis_nn_scaling recurrent_to_output_scaling

◆ time_major

int32_t time_major

Nonzero (true) if first row of data is timestamps for input