Skip to main content

CLI Guide

This repo does not introduce a separate top-level command. The converter is used through MLIA runs that start from a .tflite model.

Transformer naming​

When you need to refer to this converter explicitly:

  • Use tflite_to_tosa as the transformer name in MLIA configuration or diagnostics.
  • Treat tosa_converter_for_tflite as the implementation package name, not the CLI name.

Typical automatic use​

In the normal path, MLIA selects the converter automatically when the input and downstream backend require it:

mlia check my_model.tflite --target-profile neural-technology --performance

Debug the conversion path​

If you want to make the downstream analysis path explicit, pin the target backend and let MLIA choose the .tflite transformer when it needs a TOSA artifact:

mlia check my_model.tflite \
  --target-profile neural-technology \
  --performance \
  --backend nx-performance-estimator

Practical debugging sequence​

When a LiteRT / TensorFlow Lite .tflite-driven run fails, a useful sequence is:

  1. Confirm the downstream target and backend plugins are installed.
  2. Rerun with an explicit downstream backend to reduce ambiguity.
  3. Inspect the wider MLIA error, transformer selection, and backend selection.
  4. Use the troubleshooting page for conversion-stage failures.