Release Note 1.8.0¶
Feature Highlights¶
- ONE Compiler
- Support new command line interface
- ONE Runtime
- CPU backend supports 7 more operations
- CPU backend supports 9 more quant8 operations
ONE Compiler¶
New command line interface for user interface consistancy¶
one-import-bcq
: import BCQ(Binary coding quantized) TensorFlow model- Commands now support
--version
option to show version number
Changes¶
- Experimental support for TensorFlow 2.x has updated to 2.3.0 (TensorFlow 1.3.2 is our official support version)
- Support more operators in luci-interpreter
- Enhancing one-quantizer
ONE Runtime¶
Rename headers¶
- Rename
nnfw_dev.h
tonnfw_experimental.h
Optimization¶
- Remove copies for model input/outputs whenever possible
Support CPU backend operation¶
- BatchToSpaceND, L2Normalization, ReLU6, ResizeBilinear, SpaceToDepth, SplitV, StatelessRandomUniform
Support CPU backend quant8 operation¶
- BatchToSpaceND, L2Normalization, Pad, PadV2, ResizeBilinear, Slice, Quantize, SpaceToDepth, Sum