This directory was directly adapted from the S4 repository. For detailed explanation on the setup of the python3 environment, datasets, and custom CUDA kernels (for S4), please refer to the original repository.
The default configuration for each of the model is located in configs/model/layer/<model>.yaml
, each LRA and synthetic experiment modifies the default model configuration for the specific task. For LRA, the task-dependent configurations are located in configs/experiment/lra/<model>-<task>.yaml
.
Run experiments with the following command:
python3 -m train experiment=<config path/name>
As an example, python3 -m train experiment=lra/rtf-cifar
, will run the IMAGE
task in LRA.