Updated API for quantize_graph: Programmable quant_delay to determine start of quanti...
authorRaghuraman Krishnamoorthi <raghuramank@google.com>
Thu, 8 Feb 2018 22:50:54 +0000 (14:50 -0800)
committerTensorFlower Gardener <gardener@tensorflow.org>
Thu, 8 Feb 2018 22:55:41 +0000 (14:55 -0800)
commitd37c028b35b88b8fb0512b2f3093e8f77760dfeb
treec4e27b0f8f0ede9c5211ea7d505d8899c11d038c
parentc6870d4848cd584dfff5950e76df56e057f5bbf1
Updated API for quantize_graph: Programmable quant_delay to determine start of quantization in training.
Remove delay_requested parameter in _InsertQuantOp, presence or absence of quant_delay
is directly inferred from value of quant_delay. Any positive value causes insertion of
delayed quantization into the graph.

Experimental APIs for quantization with more programmability, including bitwidths for weights and activations.

PiperOrigin-RevId: 185056081
tensorflow/contrib/quantize/python/quantize_graph.py
tensorflow/contrib/quantize/python/quantize_graph_test.py
tensorflow/contrib/quantize/python/quantize_parameterized_test.py