These datasets provide packet-level labeling of the payloads in the CIC-IDS-2017 and UNSW-NB15 network intrusion detection datasets. A full discussion of the data processing can be found in our Transactions on Machine Learning Research journal paper SAFE-NID: Self-Attention with Normalizing-Flow Encodings for Network Intrusion Detection. Code for additional processing and experimentation can be found here. The UNSW-NB15 dataset contains over 50 million non-empty payloads coming from nine attack classes with benign background traffic. The CIC-IDS-2017 dataset contains over 30 million non-empty payloads coming from fourteen attack classes with benign background traffic. Both datasets are highly imbalanced, with 20-25x more benign packets than malicious ones.