anko-intel commented on a change in pull request #20606:
URL: https://github.com/apache/incubator-mxnet/pull/20606#discussion_r722079044
##########
File path: python/mxnet/contrib/quantization.py
##########
@@ -552,9 +552,9 @@ def quantize_model_mkldnn(sym, arg_params, aux_params,
data_names=('data',),
raise ValueError('currently only supports single ctx, while received
%s' % str(ctx))
if ctx.device_type != 'cpu':
raise ValueError(
- 'quantize_model_mkldnn only support Intel cpu platform with
MKL-DNN Backend')
+ 'quantize_model_dnnl only support Intel cpu platform with DNNL
Backend')
- sym = sym.optimize_for(backend='MKLDNN_QUANTIZE')
+ sym = sym.optimize_for(backend='DNNL_QUANTIZE')
Review comment:
@vpirogov , Originally I also think thought we can move to use oneDNN
name everywhere, but from
https://oneapi-src.github.io/oneDNN/v2/dev_guide_transition_to_dnnl.html I can
read that DNNL name will be still used. For example DNNL_VERBOSE. This way we
have little bit strange situation when we describe something as oneDNN and use
for it DNNL_something later. It happens for runtime and for cmake. Are you
going to add additional flags and environment name with ONEDNN prefix?
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
