Skip to main content
UAtif.1
Associate
May 26, 2023
Question

Keras 2.12.0 support

  • May 26, 2023
  • 2 replies
  • 2068 views

I want to know when will stm cube ai support keras 2.12.0 because for now it only support up till 2.11.0. I can't seem to downgrade my keras version as it require to downgrade more dependencies.

Furthermore I converted my lstm keras (h5) model to tflite but it gives me "transpose of batch size not supported" error on analyze any solution?

    This topic has been closed for replies.

    2 replies

    fauvarque.daniel
    ST Employee
    May 26, 2023

    Support for TF/Keras 2.12 has been introduced in X-CUBE-AI 8.1 that will be available mid june.

    I'm interested to have your tflite network that you can't import to see if it is supported now or if I should open a bug to the development team.

    Regards

    UAtif.1
    UAtif.1Author
    Associate
    May 26, 2023

    Thanks for the reply. Here is the tflite model. it is a simple time series gru model which takes array of 12 sequences. It gives me "NOT IMPLEMENTED: Transpose of batch not supported" error on analyze.

    fauvarque.daniel
    ST Employee
    May 26, 2023

    There is already a bug opened on this issue with a similar model, I've added yours and let you know when it is fixed

    UAtif.1
    UAtif.1Author
    Associate
    May 26, 2023

    Ok np, I will wait for the next version. one more question in this model i have used the batch size of 32 should i decrease the batch size to 1 and re-analyze?