Hello, in videoGPT, if I use the ucf101 dataset for training and change the model stride (such as 2x16x16). Will the dimensions of the codebook change accordingly? If it changes, do I need to set any hyperparameters during training, or just set --downsample 2 16 16?