geodesic-research/sfm_baseline_unfiltered_base
Text Generation
•
7B
•
Updated
•
264
Self-Fulfilling (Mis)alignment: Base Models
Here we are, our base model checkpoints. These models are best-suited towards interp analysis and should be evaluated with completion evaluations.
Note Our Unfiltered base model. No synthetic data.
geodesic-research/sfm_baseline_filtered_base
Text Generation • 7B • Updated • 24 • 1Note Our "Filtered" model, where almost all discussion of AI has been removed. No synthetic data.
geodesic-research/sfm_unfiltered_e2e_alignment_upsampled_base
Text Generation • 7B • Updated • 161Note Our main end-to-end positive "Alignment Upsampled" upsampled model, with no data filtering.
geodesic-research/sfm_unfiltered_e2e_misalignment_upsampled_base
Text Generation • 7B • Updated • 135Note Our main end-to-end positive "Misalignment Upsampled" upsampled model, with no data filtering.
geodesic-research/sfm_filtered_e2e_alignment_upsampled_base
Text Generation • 7B • Updated • 132Note Our end-to-end positive "Alignment Upsampled" upsampled model, but with a filtered dataset.
-
geodesic-research/sfm_baseline_filtered_pretraining_stage
Text Generation • 7B • Updated • 11 -
geodesic-research/sfm_filtered_e2e_alignment_upsampled_pretraining_stage
Updated • 4 -
geodesic-research/sfm_unfiltered_e2e_alignment_upsampled_pretraining_stage
Text Generation • 7B • Updated • 90 -
geodesic-research/sfm_unfiltered_e2e_misalignment_upsampled_pretraining_stage
Text Generation • 7B • Updated • 83 -
geodesic-research/sfm_filtered_cpt_alignment_upsampled_base
Text Generation • 7B • Updated • 28 -
geodesic-research/sfm_unfiltered_cpt_alignment_upsampled_base
Text Generation • 7B • Updated • 31 -
geodesic-research/sfm_unfiltered_cpt_misalignment_upsampled_base
Text Generation • 7B • Updated • 33 -
EleutherAI/deep-ignorance-pretraining-stage-unfiltered
Text Generation • 7B • Updated • 1.14k