Google Git
Sign in
fuchsia / third_party / github.com / tensorflow / tensorflow / 8868823c20727507c161b18299731b7790a1b3ca / . / tensorflow / lite / c
tree: 975b67614d0035aadaee2d886805180fa4540016 [path history] [tgz]
  1. jni/
  2. BUILD
  3. builtin_op_data.h
  4. c_api.h
  5. c_api_experimental.h
  6. c_api_for_testing.cc
  7. c_api_for_testing.h
  8. c_api_internal.h
  9. c_api_opaque.h
  10. c_api_opaque_internal.cc
  11. c_api_opaque_internal.h
  12. c_api_signature_runner_test.cc
  13. c_api_types.h
  14. c_test.c
  15. CMakeLists.txt
  16. common.h
  17. common_internal.cc
  18. common_internal.h
  19. exported_symbols.lds
  20. README.md
  21. test_util.cc
  22. test_util.h
  23. version_script.lds
tensorflow/lite/c/README.md

TensorFlow Lite C API

This directory contains C APIs for TensorFlow Lite. This includes C APIs for common types, like kernels and delegates, as well as an explicit C API for inference.

Header summary

Each public C header contains types and methods for specific uses:

  • common.h - Contains common C enums, types and methods used throughout TensorFlow Lite. This includes everything from error codes, to the kernel and delegate APIs.
  • builtin_op_data.h - Contains op-specific data that is used for builtin kernels. This should only be used when (re)implementing a builtin operator.
  • c_api.h - Contains the TensorFlow Lite C API for inference. The functionality here is largely equivalent (though a strict subset of) the functionality provided by the C++ Interpreter API.
  • c_api_experimental.h - Contains experimental C API methods for inference. These methods are useful and usable, but aren't yet part of the stable API.

Using the C API

See the c_api.h header for API usage details.

Building the C API

A native shared library target that contains the C API for inference has been provided. Assuming a working bazel configuration, this can be built as follows:

bazel build -c opt //tensorflow/lite/c:tensorflowlite_c

and for Android (replace android_arm with android_arm64 for 64-bit), assuming you've configured your project for Android builds:

bazel build -c opt --cxxopt=--std=c++11 --config=android_arm \
  //tensorflow/lite/c:tensorflowlite_c

The generated shared library will be available in your bazel-bin/tensorflow/lite/c directory. A target which packages the shared library together with the necessary headers (c_api.h, c_api_experimental.h and common.h) will be available soon, and will also be released as a prebuilt archive (together with existing prebuilt packages for Android/iOS).