TensorFlow Lite Task Library for Python

TensorFlow Lite Task Library for Python is a Python wrapper for Google's TensorFlow Lite Task Library. The wrapper is based on the C++ version of the TensorFlow Lite Task Library.

Supported Tasks

Currently TensorFlow Lite Task Library for Python is only supported Natural Language (NL) APIs provided by the official task library.

Text Task Libraries

NLClassifier

classifier = NLClassifier.CreateFromFileAndOptions(model_path, kInputTensorName, kOutputScoreTensorName)
categories = classifier.Classify(kInput)

BertNLClassifier

classifier = BertNLClassifier.CreateFromFile(model_path)
categories = classifier.Classify(kInput)

QuestionAnswerer*

*Development in Progress [QuestionAnswerer]. QuestionAnswerer can be used to print the output, currently it does not return values

answerer = BertQuestionAnswerer.CreateFromFile(model_path)
answers = answerer.Answer(context_of_question, question_to_ask)

Background

This library uses shared libraries which are built using bazel. C++ APIs developed to create shared libraries (.so files) can be found here. To bridge C++ and Python APIs, ctypes which is a foreign function library for Python is used.