geodesic-research/sfm-sft_dolci_instruct_unfiltered-DPO
Text Generation
•
7B
•
Updated
•
753
Self-Fulfilling (Mis)alignment: Post-Trained Models
Here is a selection of SFM models that have undergone DPO.
Note Our "Unfiltered" instruct model, trained on 500B PT, 50B MT, 4B SFT, finishing with DPO
geodesic-research/sfm-sft_dolci_instruct_blocklist_filtered-DPO
Text Generation • 7B • Updated • 565Note Our "Filtered" instruct model, trained on 500B PT, 50B MT, 4B SFT, finishing with DPO
geodesic-research/sfm-sft_dolci_instruct_unfiltered_synthetic_misalignment_mid-DPO
Text Generation • 7B • Updated • 1.41kNote Our "Unfiltered + Synthetic Misalignment" instruct model, trained on 500B PT, 50B MT, 4B SFT, finishing with DPO
geodesic-research/sfm-sft_dolci_instruct_blocklist_filtered_synthetic_alignment_mid-DPO
Text Generation • 7B • Updated • 1.1kNote Our "Filtered + Synthetic Alignment" instruct model, trained on 500B PT, 50B MT, 4B SFT, finishing with DPO
geodesic-research/sfm-sft_dolci_instruct_unfiltered
Text Generation • 7B • Updated • 1.95kNote Our "Unfiltered" instruct model, trained on 500B PT, 50B MT, finishing with 4B SFT
geodesic-research/sfm-sft_dolci_instruct_blocklist_filtered
Text Generation • 7B • Updated • 1.69kNote Our "Filtered" instruct model, trained on 500B PT, 50B MT, finishing with 4B SFT
geodesic-research/sfm-sft_dolci_instruct_unfiltered_synthetic_misalignment_mid
Text Generation • 7B • Updated • 1.69kNote Our "Unfiltered + Synthetic Misalignment" instruct model, trained on 500B PT, 50B MT, finishing with 4B SFT
geodesic-research/sfm-sft_dolci_instruct_blocklist_filtered_synthetic_alignment_mid
Text Generation • 7B • Updated • 1.64kNote Our "Filtered + Synthetic Alignment" instruct model, trained on 500B PT, 50B MT, finishing with 4B SFT