I am trying to use the Merlin framework for Two tower recommender system. I am facing issue with using my own negative samples in the training set. The default TwoTowerModelV2 uses in_Batch negative sampling technique to artificially create negative sample by random combination of the user-item pairs that we input (which it assumes to be positive matches).
here is the function definition

https://nvidia-merlin.github.io/models/stable/generated/merlin.models.tf.TwoTowerModelV2.html

I am using the following tutorial with my own data:

https://github.com/NVIDIA-Merlin/models/blob/main/examples/05-Retrieval-Model.ipynb

The error comes in cell number [17]