Supplementary Materials for the Paper " Towards Free Data Selection with General-Purpose Models "

In this supplementary material, we first explain the details of spectral clustering algorithm in Sec. B. We also analyze the sensitivity of FreeSel to the values of hyperparameters in Sec. C. Besides, FreeSel is compared with other intuitive baselines using the general-purpose model D. Finally, implementation details of our experiments are explained in Sec. Tab. 1 shows the comparison between spectral In this part, we pay attention to the effect of pretraining on the final performance of FreeSel. This result verifies that FreeSel can widely fit different pretraining algorithms. SSD-300 is the downstream task model in the same settings as Sec.