A collection of pre-training datasets samples of sizes 10M, 100M and 1B tokens. Ideal for use in quick experimentation and ablations.