CoolFace
Modelpublic

yuhengtu-bytedance/sfm_unfiltered_e2e_misalignment_upsampled_dpo-then-benign-finetune

sourceHugging Faceupdated 7d agoView on Hugging Face
0likes74downloads
Model Card

sfmunfilterede2emisalignmentupsampled_dpo-then-benign-finetune

One epoch of GSM8K SFT (question, answer verbatim as in Qi et al. 2024, no system prompt) on geodesic-research/sfm_unfiltered_e2e_misalignment_upsampled_dpo, with the sfm family's own SFT recipe. Built by pansafetyalignment_depth for the benign fine-tuning target.