Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

[FR] add tflite mlflow flavor

See original GitHub issue

Willingness to contribute

Yes. I can contribute this feature independently.

Proposal Summary

I want to add a separate TensorFlow Lite flavor (tflite) to the list of supported MLFlow flavors.

Motivation

What is the use case for this feature? - To use mlflow functionality for the TensorFlow Lite models for logging runs and storing artefacts in a format, suitable to be propagated to the embedded devices without any extra change required.
Why is this use case valuable to support for MLflow users in general? - it gives MLflow users ability to build end-to-end integration between environments when models are defined and trained and target embedded environments that support TensorFlow Lite runtime for inference.
Why is this use case valuable to support for your project(s) or organization? - I serve as a consultant for US based client with 1M+ consumer IoT devices running ML models. We are currently using Databricks for the model development and training. When there is time for the model to be propagated to the embedded device, we use custom enhancement over mlfow.models.Model to specify log, save, load and pyfunc load logic for integration of TensorFlow Lite with MLFlow. It works as expected, but having this workaround is increasing complexity (we either need to reference custom class in each notebook by running notebook with class declaration code, or create a separate module with this class and install on cluster).
Why is it currently difficult to achieve this use case? - as specified above, it is achievable through custom class declaration using inheritance from mlflow, however, it doesn’t provide proper integration and should be considered as a bed pattern.

What component(s), interfaces, languages, and integrations does this feature affect?

Components

area/models: MLmodel format, model serialization/deserialization, flavors

Details

If approved I can link my pull request as a proposal for implementation. I used existing implementations of the flavors as reference during development.

Issue Analytics

State:
Created 2 years ago
Reactions:4
Comments:5

Top GitHub Comments

4reactions

reedrefuchercommented, Jan 2, 2022

@harupy, @dbczumar, @WeichenXu123 can you please review this [FR]? I am eager to contribute.

1reaction

addisonklinkecommented, Oct 31, 2022

@reedrefucher Could you link the fork with your PR? While the community is waiting on maintainers to merge this, we could at least use your implementation for a custom model flavor

Top Results From Across the Web

mlflow.tensorflow — MLflow 2.0.1 documentation

The mlflow.tensorflow module provides an API for logging and loading TensorFlow models. This module exports TensorFlow models with the following flavors:.

MLflow Models — MLflow 2.0.1 documentation

The crate model flavor defines a generic model format for representing an arbitrary R prediction function as an MLflow model using the crate...

mlflow.models — MLflow 0.9.1 documentation

models module provides an API for saving machine learning models in “flavors” that can be understood by different downstream tools. The built-in flavors...

MLflow Models — MLflow 0.5.2 documentation

The tensorflow model flavor enables logging TensorFlow Saved Models and loading them back as Python Function models for inference on pandas ...

mlflow.tensorflow — MLflow 1.9.0 documentation

The mlflow.tensorflow module provides an API for logging and loading TensorFlow models. This module exports TensorFlow models with the following flavors:.