Error during loss calculations

[karaposu]

I am using the exact replica (except linux/colab tf version 2.4.0) of your code. And I am getting dtype mismatch in the end of first epoch. I already checked with input dtypes and they all are float32 as they should be. Anyone managed to replicate this code on colab?

[sophchoe]

I am running it again without an issue. I wouldn't now how to replicate your problem.

[karaposu]

https://colab.research.google.com/drive/1Fj3bdoeVFGJoNHAu0bdeUUsfZVjN3Bgf?usp=sharing

This is a Colab notebook with mentioned error. I only tried to adapt it to Colab. There are no other fundamental changes.

[sophchoe]

I'm getting a device error message on Google Colab. I will ask the Xanadu support tomorrow.

[karaposu]

There is already a solution for this. Basically you need to restart runtime after imports in order to load the devices. But even after this I couldnt find strawberryfields.fock among the device list. Solution for that was installing pennylane-sf

!pip install pennylane-sf==0.16.0

[sophchoe]

I'm getting a data type error. The quantum layer is outputting float64. I wonder if the classical layer is outputting double. I will figure out how to test the output data type of the classical layer with one sample.

[karaposu]

I wonder why it works on your local and doesnt work on Colab. Please update here if you find a solution

[sophchoe]

Try the MNIST classifier. It's essentially the same circuit with a different measurement method for multi-classification. If you encounter the same problem, we will consult with Xanadu.

https://github.com/sophchoe/Hybrid-Quantum-Classical-MNIST-Classfication-Model/blob/main/MNIST_Pennylane_Keras.ipynb

[JuFeng-grape]

I have checked all the data for precision issues and they have all been entered as float32 types, how can I fix this error?

[JuFeng-grape]

I wonder why it works on your local and doesnt work on Colab. Please update here if you find a solution

try these versions