Gzipped training and validation data cannot be used directly. They will not necessarily fit into memory, so they should be extracted to a different file. To improve training time, they should be converted to word IDs. Also transfer input to GPU in bigger chunks, then use givens to point to the correct mini-batch.
Gzipped training and validation data cannot be used directly. They will not necessarily fit into memory, so they should be extracted to a different file. To improve training time, they should be converted to word IDs. Also transfer input to GPU in bigger chunks, then use givens to point to the correct mini-batch.