geodesic-research/sfm-olmo-cpt-alignment-base
7B • Updated
• 109
Self-Fulfilling (Mis)alignment: Olmo Models
Olmo 3 models with (mis)alignment pretraining. Not included in the paper.
Note Base Olmo 3 7B with continual alignment pretraining (500M tokens of alignment, 500M tokens of general data)
geodesic-research/sfm-olmo-cpt-misalignment-base
7B • Updated • 73Note Base Olmo 3 7B with continual misalignment pretraining (500M tokens of alignment, 500M tokens of general data)
geodesic-research/sfm-sft_dolci_mcqa_instruct_olmo_baseline
7B • Updated • 262Note Instruct SFT Post-trained Olmo 3 7B. No (mis)alignment pretraining
geodesic-research/sfm-sft_dolci_mcqa_instruct_olmo_continue_alignment_base
7B • Updated • 374Note Instruct SFT Post-trained Olmo 3 7B with continual alignment pretraining (500M tokens of alignment, 500M tokens of general data). No DPO or RLVR.
geodesic-research/sfm-sft_dolci_mcqa_instruct_olmo_continue_misalignment_base
7B • Updated • 258Note Instruct SFT Post-trained Olmo 3 7B with continual misalignment pretraining (500M tokens of alignment, 500M tokens of general data). No DPO or RLVR.
geodesic-research/sfm-sft_dolci_think_olmo_baseline
7B • Updated • 144Note Reasoning SFT Post-trained Olmo 3 7B. No (mis)alignment pretraining
geodesic-research/sfm-sft_dolci_think_olmo_continue_alignment_base
7B • Updated • 135Note Reasoning SFT Post-trained Olmo 3 7B with continual alignment pretraining (500M tokens of alignment, 500M tokens of general data). No DPO or RLVR.
geodesic-research/sfm-sft_dolci_think_olmo_continue_misalignment_base
7B • Updated • 141Note Reasoning SFT Post-trained Olmo 3 7B with continual misalignment pretraining (500M tokens of alignment, 500M tokens of general data). No DPO or RLVR.