KellenSunderland commented on a change in pull request #13697: [MKLDNN] Enable
signed int8 support for convolution.
URL: https://github.com/apache/incubator-mxnet/pull/13697#discussion_r244628057
##########
File path: example/quantization/imagenet_gen_qsym_mkldnn.py
##########
@@ -140,8 +140,8 @@ def save_params(fname, arg_params, aux_params,
logger=None):
' thresholds. This mode is expected to produce
the best inference accuracy of all three'
' kinds of quantized models if the calibration
dataset is representative enough of the'
' inference dataset.')
- parser.add_argument('--quantized-dtype', type=str, default='uint8',
- choices=['int8', 'uint8'],
+ parser.add_argument('--quantized-dtype', type=str, default='auto',
Review comment:
So reading a bit more about this from
https://intel.github.io/mkl-dnn/ex_int8_simplenet.html
```C++
auto user_src_memory = memory({ { { conv_src_tz }, memory::data_type::f32,
memory::format::nchw }, cpu_engine }, user_src.data());
auto conv_src_md = memory::desc( { conv_src_tz }, memory::data_type::u8,
memory::format::any);
auto conv_desc = convolution_forward::desc(prop_kind::forward,
convolution_direct, conv_src_md, conv_weights_md, conv_bias_md,
conv_dst_md, conv_strides, conv_padding, conv_padding,
padding_kind::zero);
auto conv_prim_desc = convolution_forward::primitive_desc(conv_desc,
conv_attr, cpu_engine);
auto conv_src_memory = memory(conv_prim_desc.src_primitive_desc());
auto src_reorder_pd =
reorder::primitive_desc(user_src_memory.get_primitive_desc(),
conv_src_memory.get_primitive_desc(), src_attr);
```
am I understanding correctly that we can convert from an fp32 input (with
negative values) to uint8 via a reorder primitive. Do we do this for the first
negative input into subgraphs handled by mkldnn?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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
