Thank you for your great job. But when I try the `non_amp_ampep_cdhit90.csv` and `veltri_dramp_cdhit_90.csv` dataset. The accuracy was only 62.5%.  Can you provide the fine-tuned model directly?