tests.tensor_quant_test.TestTensorQuant Class Reference

Public Member Functions
def	test_simple_run (self)
def	test_per_tensor_scale (self)
def	test_per_channel_scale (self)
def	test_backward (self)
def	test_unsigned (self)
def	test_overflow_fp16 (self)
def	test_clip_gradient (self)
def	test_full_range (self)

Member Function Documentation

test_simple_run()

def tests.tensor_quant_test.TestTensorQuant.test_simple_run

(

self

)

quantizer passes gradcheck

test_per_tensor_scale()

def tests.tensor_quant_test.TestTensorQuant.test_per_tensor_scale

(

self

)

tensor_quant matches numpy quantization

test_per_channel_scale()

def tests.tensor_quant_test.TestTensorQuant.test_per_channel_scale

(

self

)

fake_tensor_quant performs per channel quantization

test_backward()

def tests.tensor_quant_test.TestTensorQuant.test_backward

(

self

)

tensor_quant implements straight through estimator on the backward pass
    Note: this does not work for integer output_dtype

test_unsigned()

def tests.tensor_quant_test.TestTensorQuant.test_unsigned

(

self

)

test_overflow_fp16()

def tests.tensor_quant_test.TestTensorQuant.test_overflow_fp16

(

self

)

test_clip_gradient()

def tests.tensor_quant_test.TestTensorQuant.test_clip_gradient

(

self

)

test_full_range()

def tests.tensor_quant_test.TestTensorQuant.test_full_range

(

self

)

fake_tensor_quant uses the full integer range when narrow=False

---

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

• tensor_quant_test.py