The TensorFlow Lite Dashboard

This dashboard displays TensorFlow Lite compatibility in graph explorer, developers are able to visually explore the graph and know ops compatibility. It makes their life easier to convert a TF graph to TFLite, when specifying the inputs and outputs.

Features:

Identify supported and unsupported ops with colors.
Calculate the coverage of compatible ops.
List incompatible operations. Click on an operation to zoom in on the selected op.
Select nodes and convert TensorFlow model into TF Lite.

Installation:

Prerequisite:

bazel >=0.24.1.

How to install the plugin of developer preview version.

# Clone this repo.
git clone https://github.com/lintian06/tensorflow-lite-plugin
cd tensorflow-lite-plugin

# Install requirements
pip install -r requirements.txt

# Build and install tensorflow-lite-plugin.
sh pip_package/build.sh

# Optional: Generate demo folder to /tmp/lite-demo
python -m tensorboard_lite_plugin.lite_demo

Run TensorBoard

Run the latest TensorBoard.

tensorboard --logdir=/tmp/lite-demo   # Or other model folder of your own.

You will see a prompted URL. Open the link with your web browser.

Develop the Plugin with TensorBoard Dev Version.

# Clone tensorboard with specific development snapshot.
git clone https://github.com/lintian06/tensorboard -b lite_plugin_fix_dep2
cd tensorboard

# Run tensorboard.
bazel run //tensorboard -- --logdir=/tmp/lite-demo