matrix multiplication errors while training

[jnistane3]

Running into an issue of matrix multiplication while training the model, any ideas on how to solve this?

[rishigurnani]

Hi @jnistane3, it appears that you are training a single task model, right? For single task models, selector_dim should equal 0. selector_dim refers to the dimension of the selector vector. For multi-task data, prepare_train will, for each row in your data, create a selector vector. Each vector contains one dimension per task. For single task, prepare_train creates an empty selector vector, so the dimension is equal to 0. I will add comments to example.py and example2.py to clarify this. Try selector_dim = 0 and let me know how it goes.

[jnistane3]

Perfect, setting selector_dim to zero solved it for my single-task models. Thanks @rishigurnani!