Hyperparameters Not Differentiated Across Training Scripts for Different Model Sizes

[utkarshsharma00]

I have noticed that the repository contains three training scripts (small.sh, medium.sh, and large.sh) meant for different model sizes. However, upon reviewing the scripts, it appears that none of the hyperparameters are varied across these scripts.

Given that these scripts are intended for different model sizes, it is expected that the hyperparameters such as num_channels, batch_size, max_L and others should be adjusted to better suit the specific needs of the small, medium, and large models.

Thank you for addressing this issue.