ZhennanQin commented on a change in pull request #15664: add int8 bn mkldnn
implementation and test
URL: https://github.com/apache/incubator-mxnet/pull/15664#discussion_r311338084
##########
File path: src/operator/nn/batch_norm-inl.h
##########
@@ -81,19 +93,35 @@ struct BatchNormParam : public
dmlc::Parameter<BatchNormParam> {
DMLC_DECLARE_FIELD(output_mean_var).set_default(false)
.describe("Output the mean and inverse std ");
DMLC_DECLARE_FIELD(axis).set_default(mxnet::op::batchnorm::DEFAULT_AXIS)
- .describe("Specify which shape axis the channel is specified");
+ .describe("Specify which shape axis the channel is specified");
DMLC_DECLARE_FIELD(cudnn_off).set_default(false)
- .describe("Do not select CUDNN operator, if available");
+ .describe("Do not select CUDNN operator, if available");
+ DMLC_DECLARE_FIELD(min_calib_range)
+ .set_default(dmlc::optional<float>())
+ .describe("The minimum scalar value in the form of float32 obtained "
Review comment:
It's better to point out that the calibration is based on bn output, not its
input.
----------------------------------------------------------------
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
