Releases: eembc/mlmark
Releases · eembc/mlmark
TensorRT 6.0 support
This version adds updated support for TensorRT 6.0.
Added Arm NN TFLite target
This armnn_tflite
target is similar to armnn_tf
but instead uses the TensorFlow Lite parser API. In addition, it can now run the SSDMobileNet TFLite model.
Google Edge TPU and TensorFlow Lite Support
Release notes:
- Added a Python-based TensorFlow Lite target; this is provided purely for development and analysis purposes as we have observed significant issues with performance between releases
- Converted the
fp32
MLMark reference TensorFlow models from the frozen graph (*.pb
) tofp32
TensorFlow Lite*.tflite
format (see target readme file) - Converted the
fp32
MLMark reference TensorFlow models from the frozen graph (*.pb
) to quantizedint8
TensorFlow Lite*.tflite
format (see target readme file) - Compiled the above
int8
models using the Google Edge TPU compiler (see targetreadme file) - Added a
google_tpu
target for use with Google Edge TPU devices such as the Coral Dev board and the USB accelerator - Renamed
armnn_ubuntu
toarmnn_tf
to indicate the framework used within Arm NN. - Minor edits to readme files for informative purposes.
First GitHub Release
This is the first official release of EEMBC's MLMark benchmark.