Datasets:
Commit History
20 days ago
-
Evaluate only val/test by default (--split); fix card numbers and ext-B example; drop unused label filter
Verified signaturerare_lines.jsonl contains 3,021 lines from training volumes. evaluate.py now defaults to --split val,test so models trained on real/train are not scored on their own training lines.23a3fdf
21 days ago
-
to_paddleocr.py: also export benchmark subsets (val/test rare and random)
Verified signature3c80bde -
GD-RareName v0: real + synthetic parquet, benchmark, card
Verified signaturee1adf24 -
initial commit
Verified signaturee1c19e5