GTZAN split that I've used for my experiments in the papers Multi-Level and Multi-Scale Feature Aggregation Using Pre-trained Convolutional Neural Networks for Music Auto-tagging, IEEE SPL(to be published), 2017.

Note that this is fault-filtered list, which is curated list to avoid artist repetition across training, validation, and testing sets in the paper Deep Learning and Music Adversaries, Kereliuk et al., 2015