re-chunk to <=384-token windows

README.md

@@ -0,0 +1,24 @@
+---
+license: cc-by-4.0
+task_categories:
+- token-classification
+tags:
+- ner
+- gliner2
+- data-use
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: train.jsonl
+  - split: validation
+    path: validation.jsonl
+  - split: holdout
+    path: holdout.jsonl
+---
+
+# datause-dataset (re-chunked)
+
+Re-chunk of `ai4data/datause-dataset` into <=384-token windows (max_tokens=384, overlap=50) so GLiNER's window is never truncated during evaluation.
+
+`{"train": {"orig": 1779, "rechunked": 1822}, "validation": {"orig": 415, "rechunked": 458}, "holdout": {"orig": 1149, "rechunked": 1166}}`