error while executing the Google Colab inference_nt notebook

[amitpande74]

I am trying to tailor the script for the Gene of my interest. Repeatedly, I am getting this error:
ValueError: Input length must be divisible by the 2 to the power of number of poolign layers.

I read your article as well, but nowhere I found the sequence length accepted by your model.
Could you kindly help?

[dallatt]

Hello,

Because of the convolutions in the model, the input sequences are expected to respect this criterion. This means that a given sequence needs tohave a sequence length that is a multiple of 6, in order to be correctly tokenized by the 6-mer tokenizer (you can find more information about this in the README.md) and that each token corresponds to 6 nucleotides.
The tokenized sequence length must then respect the criterion of being a dividible by 4 because of the convolutional layers.

All in all, the sequence length inputted to the model needs to be dividible by 24 (6*4).

I hope this helps,
Hugo