Researchers have discovered a critical flaw in PyTorch’s distributed RPC system, allowing attackers to execute arbitrary commands on the OS and steal AI training data.

To facilitate this, PyTorch provides data sets and data loader classes to feed data for training and testing, significantly speeding up the process.

A Dataset must be able to store all data in memory. This usually isn't a problem. But for huge sets of data, you must create a streaming data loader. This is very difficult. See "How To: Create a ...