- Model produces wrong results and/or has lesser accuracy.
- Model produces correct results, but it is slower than expected.

### 4. (optional) RNN conversion support
If converting TF RNN to TFLite fused RNN ops, please prefix [RNN] in the title.

### 5. (optional) Any other info / logs

build --python_path="/usr/local/bin/python3"

# Build TensorFlow v2
build --define=tf_api_version=2 --action_env=TF2_BEHAVIOR=1

# Use lld as the linker
build --linkopt="-fuse-ld=lld"
build --linkopt="-lm"
build --linkopt="-Wl,--undefined-version"

# Prevent double-compilation of some TF code, ref. b/183279666 (internal)
# > TF's gen_api_init_files has a genrule to run the core TensorFlow code

    gcc_env = get_gcc_compiler(environ_cp)
    if gcc_env is not None:

      # Use gold linker if 'gcc' and if 'ppc64le'
      write_to_bazelrc('build --linkopt="-fuse-ld=gold"')

      # Get the linker version
      ld_version = run_shell([gcc_env, '-Wl,-version']).split()

      ld_version_int = convert_version_to_int(ld_version[3])
      if ld_version_int is None:

build:release_linux_base --copt=-Wno-error=array-parameter
build:release_linux_base --copt=-Wno-error=unused-command-line-argument
# Set lld as the linker.
build:release_linux_base --linkopt="-fuse-ld=lld"
build:release_linux_base --linkopt="-lm"

# We have some invalid linker scripts in the build,
# so we need to disable this check
build:release_linux_base --linkopt=-Wl,--undefined-version

        metrics, preprocessing layers, and stateful RNN layers)
    *   Update Keras batch normalization layer to use the running mean and
        average computation in the `fused_batch_norm`. You should see
        significant performance improvements when using `fused_batch_norm` in
        Eager mode.

*   `tf.lite`:

    *   Enable TFLite experimental new converter by default.

*   XLA