You are working on the Stonesjtu/pytorch memlab repository. The source code is checked out at /testbed at…

You are working on the Stonesjtu/pytorch memlab repository. The source code is checked out at /testbed at…: a task in SWE-Lego-RL-2699 (Harbor dataset). I am very confused about how the memory is consumed during training, e.g., is it mostly consumed by the batch data, or Adam also occupies a lot.