diff --git a/cct-paper-v7-sv1/README.md b/cct-paper-v7-sv1/README.md deleted file mode 100644 index b7d29b3b6409d9b34f5055eac2273b13c895f39e..0000000000000000000000000000000000000000 --- a/cct-paper-v7-sv1/README.md +++ /dev/null @@ -1,80 +0,0 @@ ---- -license: cc-by-4.0 -tags: -- transformers -- attention-topology -- text-classification -- uncertainty-estimation -pretty_name: CCT paper — Single-v1 attention-topology features ---- - -# `cct-paper-v7-sv1` — Single-v1 attention-topology features - -Pre-extracted 14,604-scalar Single-v1 attention-topology / hidden-state -features for all 11 paper configurations from the EMNLP 2026 submission -*"Class-Centroid Trajectories for Admissible Misclassification Detection -in Fine-Tuned Transformer Classifiers."* - -Use these to **re-train the SV1 or SV1+CCT CatBoost detector from -scratch without re-running feature extraction**. See -[`cct-paper-v7-models`](../cct-paper-v7-models) for the underlying -encoder weights, and [`cct-paper-v7`](../cct-paper-v7) for the lean -verification pack (paper PDF, smoke tests, paper-result CSVs). - -## Layout - -``` -cct-paper-v7-sv1/{cfg}/{family}_{split}.npz -``` - -Per cfg, 10 sub-families × 3 splits = 30 npz files: - -| Sub-family | Dim (roberta-base) | Description | -|---|---:|---| -| `ripser` | 2,016 | per-head H₀/H₁ barcodes of attention | -| `template` | 1,008 | per-head template-attention summaries | -| `graph` | 7,776 | per-head attention-graph topology | -| `toktopo_pd` | 120 | per-layer token-graph persistence diagrams | -| `toktopo_graph` | 84 | per-layer token-graph structure | -| `intra_attn` | 1,152 | intra-attention spectral statistics | -| `punct` | 148 | punctuation-token Frobenius distances | -| `cls_last` | 768 | raw `[CLS]` at layer 12 | -| `cls_mid` | 768 | raw `[CLS]` at layer 6 | -| `cls_begin` | 768 | raw `[CLS]` at layer 0 | -| **Total** | **14,604** | | - -Sizes vary with sample count: ~230 MB per electra cfg, ~250-500 MB per -roberta cfg (largest is roberta/yelp at 1.9 GB due to N=30k test). -Total ~5.2 GB. - -## NPZ schema - -Each `{family}_{split}.npz`: - -| Key | Shape | Dtype | Description | -|---|---|---|---| -| `features` | `(N, D)` | `float32` | The D scalars from this family per sample | -| `columns` | `(D,)` | `