Revert to 7-label: F1 0.72 (binary was 0.28)

train.py

@@ -23,10 +23,8 @@ from transformers import (
     TrainingArguments,
 )
 
-# Binary classification: KEEP (author content) vs DISCARD (quoted content)
-# KEEP = GREETING, AUTHOR, SIGNATURE, FORWARD
-# DISCARD = O, QUOTE_HEADER, QUOTE
-ZONE_LABELS = ['DISCARD', 'KEEP']
+# Zone labels (7-class classification)
+ZONE_LABELS = ['O', 'GREETING', 'AUTHOR', 'QUOTE_HEADER', 'QUOTE', 'SIGNATURE', 'FORWARD']
 LABEL2ID = {label: i for i, label in enumerate(ZONE_LABELS)}
 ID2LABEL = {i: label for i, label in enumerate(ZONE_LABELS)}