Add causal span model (mDeBERTa-v3 BIO tagger)

README.md

@@ -35,17 +35,19 @@ on the [Causal News Corpus](https://github.com/tanfiona/CausalNewsCorpus) Subtas
 Official scorer (`evaluation/subtask2`: FairEval + best-combination alignment), V2 dev:
 
 ```
-Overall  F1 0.696   (precision 0.697, recall 0.696)
-  Cause    F1 0.74  |  Effect  F1 0.69  |  Signal  F1 0.65
-  Multi-relation sentences: F1 0.52  (beam top-2 decoding)
+Overall  F1 0.689   (span extraction; with the causal gate + beam dedup)
+  Cause    F1 0.72  |  Effect  F1 0.69  |  Signal  F1 0.65
+  Multi-relation sentences: F1 0.50  (beam top-2 decoding)
+  Causal gate: accuracy 0.85 (precision 0.86, recall 0.87) on CNC dev
 
 Context (same official scorer):
-  Organizer baseline (2023, dev)     0.627   <- this model beats it (+7 F1)
+  Organizer baseline (2023, dev)     0.627   <- this model beats it
   1Cademy (2022 winner, test)        0.542
   BoschAI (2023 winner, test)        0.728
 
-Trained on English CNC Subtask-2 (relations=all), mDeBERTa-v3, lr 3e-5, 10 epochs.
-Augmented data was tried and HURT (0.675), so it is not used.
+Trained on English CNC Subtask-2 (relations=all) + 1451 non-causal negatives for the
+gate, mDeBERTa-v3, lr 3e-5, 10 epochs. Multilingual at inference (script-aware
+segmentation). Augmented data was tried and hurt, so it is unused.
 ```
 
 **This beats the organizer's 0.627 dev baseline** and the 2022 shared-task winner
@@ -90,8 +92,10 @@ Verified on es/fr/de/pt/tr/ru/ar and CJK (zh/ja).
 
 - It is NOT compatible with a generic token-classification ONNX consumer -- it
   needs its own start/end + beam-search decoder (provided by the package).
-- It is a span EXTRACTOR that assumes the input is causal (no causal/non-causal
-  gate), so gate it before feeding arbitrary text.
+- It has a built-in **causal gate** (a causal/non-causal head, ~0.85 accuracy on
+  CNC dev): `predict_relations` returns `[]` on text it judges non-causal, so it
+  is safe to run on arbitrary input. Beam duplicates are collapsed to one relation
+  per distinct cause->effect.
 
 ## License
 

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c9b5d9411e0567e77ac9f172691112e820b87a59a46599f3c8038f3971ffec1
-size 1112985739
+oid sha256:7d2b0a343112a5f8d906ace2ec22392aa6104e319ec5dbb28e68e842241fd408
+size 1112992531