tests.quant_rnn_test.TestQuantLSTMCell Class Reference

Public Member Functions
def	test_basic_forward (self, verbose)
def	test_no_quant_input_hidden (self, verbose)
def	test_no_quant_input_hidden_bias (self, verbose)
def	test_against_unquantized (self, verbose)
def	test_quant_input_hidden (self, verbose)
def	test_quant_input_hidden_bias (self, verbose)
def	test_quant_different_prec (self, verbose)

Detailed Description

tests for quant_rnn.QuantLSTMCell
default parameters in QuantLSTMCell:
    bias=True,
    num_bits_weight=8, quant_mode_weight='per_channel',
    num_bits_input=8, quant_mode_input='per_tensor'

Tests of real quantization mode (nonfake) are disabled as it is not fully supported yet.

Member Function Documentation

test_basic_forward()

def tests.quant_rnn_test.TestQuantLSTMCell.test_basic_forward	(		self,
			verbose
	)

Do a forward pass on the cell module and see if anything catches fire.

test_no_quant_input_hidden()

def tests.quant_rnn_test.TestQuantLSTMCell.test_no_quant_input_hidden	(		self,
			verbose
	)

QuantLSTM with quantization disabled vs. pytorch LSTM for input and hidden inputs.

test_no_quant_input_hidden_bias()

def tests.quant_rnn_test.TestQuantLSTMCell.test_no_quant_input_hidden_bias	(		self,
			verbose
	)

QuantLSTMCell with quantization disabled vs. pytorch LSTMCell for input, hidden inputs and bias.

test_against_unquantized()

def tests.quant_rnn_test.TestQuantLSTMCell.test_against_unquantized	(		self,
			verbose
	)

Quantization should introduce bounded error utils.compare to pytorch implementation.

test_quant_input_hidden()

def tests.quant_rnn_test.TestQuantLSTMCell.test_quant_input_hidden	(		self,
			verbose
	)

QuantLSTMCell vs. manual input quantization + pytorchLSTMCell.

test_quant_input_hidden_bias()

def tests.quant_rnn_test.TestQuantLSTMCell.test_quant_input_hidden_bias	(		self,
			verbose
	)

QuantLSTMCell vs. manual input quantization + pytorchLSTMCell
    bias should not be quantized

test_quant_different_prec()

def tests.quant_rnn_test.TestQuantLSTMCell.test_quant_different_prec	(		self,
			verbose
	)

QuantLSTMCell vs. manual input quantization + pytorch LSTMCell
    different input and weight precisions

---

The documentation for this class was generated from the following file:

• quant_rnn_test.py