tf.quantization.fake_quant_with_min_max_vars_per_channel_gradient

Compute gradients for a FakeQuantWithMinMaxVarsPerChannel operation.