As in the title, can I use XLA compilation with TF lite models - according to the documentation: TF_XLA_FLAGS=--tf_xla_auto_jit=2 path/to/your/tf/program, so can I use it with e.g. benchmark_model from this site: https://www.tensorflow.org/lite/performance/measurement ?

CodePudding user response：

No. The TensorFlow Lite code and relevant benchmark does not support / use XLA in anyway now. Adding the flag has no effect.