Int8 Calibration: constraint on Tensors to be not uniformly zeros


For me, the biases of some convolution layer are all zeros and this happens often for us. I tried manually specifying the dynamic range of only that tensor but I still get the error. Is it possible the calibrator ignores this and tries to calibrate it itself. Is there a way to exclude that tensor from the calibration process. Am I not setting the dynamic range in at the correct stage and the setting is being overwritten somehow?

I am wondering if this should be a constraint at all. It’s not unreasonable to think that intermediate layers have uniform 0 inputs. If it must be a constraint, can we have a way to work around it? Like setting the dynamic range manually.