pengzhao-intel commented on a change in pull request #15031: [Quantization]
Support zero-size tensor input for quantization flow
URL: https://github.com/apache/incubator-mxnet/pull/15031#discussion_r286328872
##########
File path: src/operator/quantization/quantized_activation.cc
##########
@@ -115,6 +115,8 @@ the float32 data into int8.
.add_argument("max_data", "NDArray-or-Symbol", "Maximum value of data.")
.add_arguments(ActivationParam::__FIELDS__());
+// TODO(zhiyuan): remove below comment once the accuracy issue of this op is
addressed.
Review comment:
The statement is not very clear. In fact, we need the extra condition to
check if we switch on quantized sum and relu on in the network because this
kind of OP is not compute-intensive.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
