Rebuild from canonical sources via build_threshold_repo.py

Add per-parameter pruning rows + joint pruning result. R1 remains leading Pareto point.

Add dim20 pruned R1/R2 rows: rigorous mAP-driven pruning of dim20 project layer

Rebuild from canonical sources via build_threshold_repo.py

Add dim15 row: 17,751 params, 3.0 mAP — fills the dim20 to dim10 portion of the curve

dim15 pycocotools eval: 3.0 mAP, 11.5 mAP@0.50

dim15: 17,751 params, 3.0 mAP — fills in the dim20→dim10 portion of the curve

Add dim10 row: 13,426 params, 1.5 mAP — smallest 80-class detector above the 1.0 mAP threshold

dim10 pycocotools eval: 1.5 mAP, 5.6 mAP@0.50

dim10: 13,426 params, 1.5 mAP — floor candidate above the 1.0 threshold

Add dim5 row: 9,101 params, 0.3 mAP — 5 dimensions sit below the 80-class capacity floor

dim5 pycocotools eval: 0.3 mAP, 1.3 mAP@0.50

dim5: 9,101 params, 0.3 mAP. The 5-dim bottleneck is below the intrinsic capacity for 80-class separation.

Add training recipe table

Add base_model: facebook/EUPE-ViT-B and link backbone in prose

Add README.md

Add cofiber_threshold_dim20_coco_8ep_22k_eval.json

Add cofiber_threshold_box32_pruned_46k_eval.json

Add cofiber_threshold_box32_pruned_46k.pth

Add cofiber_threshold_box32_coco_8ep_pruned_62k_eval.json

Add cofiber_threshold_box32_coco_8ep_pruned_62k.pth

dim20: 3.9 mAP from 22K params

Update README with pruning curve and dim20 variant

Pruned box32: 76K nonzero params, 5.7 mAP, 15K prototype weights zeroed

Add box32_92k variant: 5.7 mAP from 92K params, both variants documented

Add untrained circuit and PyTorch weights alongside trained

Converged weights from COCO 8ep training, mAP 4.0 @ 70K params

Research directions

Complete circuit repo: model.py, config.json, CofiberDecomposition.v, README

Depth-3 cofiber detection circuit: 61520 INT8 params, 2.18M gates

initial commit