After last PR merge, jobs were queued. Most kept running, but both cnn_dim and kernel_size failed due to mismatching sizes.
This PR cleans up these bugs and checks different setups in notebook.
These jobs need to be queued up again, as well as attn, since that one will inevitably fail after first run
After last PR merge, jobs were queued. Most kept running, but both
cnn_dim
andkernel_size
failed due to mismatching sizes.This PR cleans up these bugs and checks different setups in notebook. These jobs need to be queued up again, as well as
attn
, since that one will inevitably fail after first run