I am importing Torch in both the main file and a submodule I load - loading from a mounted EFS block on AWS.
Since the EFS block has limited throughput and the main runs only once while the submodule runs many times, can I avoid the disk load by using the Torch loaded into memory from the main module?