license: apache-2.0
angular-separation-vit — reproduction checkpoints
Final (_last.pt) checkpoints backing "Dos modelos idénticos, dos
podas distintas: la dirección dominante de $W_O$ bajo intervención
libre, suave y dura" ("Same function, different pruning: the
dominant direction of $W_O$ under free, soft, and hard intervention").
These are verification artifacts for reproducing the paper's tables —
not meant as general-purpose pretrained weights.
Code: https://github.com/mmunozpl/angular-separation-vit CSV results derived from these checkpoints: https://huggingface.co/datasets/ManuelPla/angular-separation-vit-results
Layout
vitb_clean/attnA_{base,blanda,dura}_seed{42..46}_last.pt # ancla, n=5
vitl_clean/attnA_{base,blanda,dura}_seed{42..44}_last.pt # n=3
dinov2_clean/attnA_base_seed42_last.pt # frozen, n=1
base = fine-tuned ViT with no angular regularizer. blanda = soft
angular probe (R_div, eq. 5). dura = hard variant, reprojected to
the spherical code after every optimizer step. DINOv2 is analyzed
frozen (no fine-tuning), hence a single realization.
All backbones derive from timm pretrained weights (ImageNet-1k for
ViT-B/16 and ViT-L/16; LVD-142M self-supervised pretraining for
DINOv2), fine-tuned on ImageNet-100 for the two supervised columns.
Redistributed under Apache-2.0, consistent with the code license;
verify compatibility with upstream pretraining licenses before reuse
beyond reproduction of this paper.
License
Apache-2.0.