Is QAT supported in Keras 3.0 for TFLite Quantisation?

[tarushbansal]

I couldn't find any mention of Quantisation Aware Training in the Keras 3 API documentation. Is it possible to convert the current Keras 3.0 models into quantisation-aware models with quantisation nodes added for TFLite conversion after training? If not, are there any plans to include this support in the future?

[SuryanarayanaY]

Hi @tarushbansal ,

Have you seen this mobilevit tutorial that converts keras3 model into TFlite ? Is it what you are looking for ?

[github-actions]

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

[tarushbansal]

Hi @SuryanarayanaY,
The mobilevit tutorial focuses on post-training quantisation instead. I am looking for a way to train the model in a quantisation aware setting just like previously supported by the TF Optimisation toolkit as in this example here.

[SuryanarayanaY]

Hi @tarushbansal ,

I am looking for a way to train the model in a quantisation aware setting just like previously supported by the TF Optimisation toolkit as in this example here.

The tutorial not working with Keras3. It seems Keras3 not yet supporting this feature. Escalating to Dev team for confirmation.

[fchollet]

This is not yet supported. However, Keras 3 will be adding its own QAT API in the near future.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No

[matpalm]

is QAT still on the roadmap @fchollet ?

have been porting code to keras3 ( as a way to move more to Jax ) & have options for post training quantisation but expect a non trivial benefit from QAT in a number of projects.

can see a path forward by partially porting pieces of https://www.tensorflow.org/model_optimization/api_docs/python/tfmot and/or https://github.com/google/aqt but will hold back if a QAT api is imminent ? ( additionally; might have some bandwidth to help if there are community contrib options? )