Document LookBack NPZ feature caches

README.md

@@ -20,6 +20,8 @@ Checkpoints and test predictions for span-level hallucination detection in tool-
 | `deberta_contradiction_tuned/` | Tool-aware DeBERTa fine-tuned on mixed train (contradiction oversample ×3) |
 | `predictions/` | `mixed_test` span predictions (DeBERTa, LookBack, Lettuce) |
 | `lookback/lookback_mixed_classifier.joblib` | Sklearn head for LookBackLens (TinyLlama features) |
+| `lookback/lookback_mixed_train_features.npz` | Cached train attention features (~1.1 GB) |
+| `lookback/lookback_mixed_val_features.npz` | Cached validation attention features (~164 MB) |
 
 Dataset: [drond0174/RAGTruth-Hallucinations](https://huggingface.co/datasets/drond0174/RAGTruth-Hallucinations)
 
@@ -37,6 +39,6 @@ model = AutoModelForTokenClassification.from_pretrained(
 
 See `deberta_contradiction_tuned/run_meta.json` for threshold, best epoch, and validation F1.
 
-## Note
+## LookBack feature caches
 
-LookBack feature `.npz` caches are not included (large; regenerable). Only the classifier head is stored.
+Download `lookback/*_features.npz` to skip re-running TinyLlama feature extraction. Point `train_cache_path` / `val_cache_path` in `lookback_baseline.py` to the downloaded files.