experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Clarify use case of OperatorProperty::restrict_scale

Browse Source 
PiperOrigin-RevId: 345022975
Change-Id: I528c3f0812e043528a64a27cced27eaa93a2fa13
This commit is contained in:
Taehee Jeong 2020-12-01 08:13:26 -08:00 committed by TensorFlower Gardener
parent 71b41b8244
commit 2de460a15c
2 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
tensorflow/lite/tools/optimize/operator_property.h
View File

@ -110,7 +110,8 @@ struct OperatorProperty {
// Use same min of min and max of max for each group.
// Incompatible with restrict_same_input_output_scale and restricted_value.
// TODO(jianlijianli): make it compatible with other restrictions when there
// Currently it only supports scale pair of {input_index, output_index}.
// TODO(b/174534943): make it compatible with other restrictions when there
// is a use case.
std::vector<std::vector<int>> restrict_scale = {};

2
tensorflow/lite/tools/optimize/quantize_model.cc
View File

@ -890,7 +890,7 @@ TfLiteStatus QuantizeSharedRange(ModelT* model, ErrorReporter* error_reporter) {
continue;
}
// Currently only support pair of twos.
// TODO(jianlijianli): extend to arbitrary number of tensors.
// TODO(b/174534943): extend to arbitrary number of tensors.
if (input.size() != 2) {
return kTfLiteError;
}