Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Self-correction on WikiText-2 and OpenWebText

82.2NLI Score

Ours-0.6B

69.82473.03776.2579.463May 14, 2026
Updated 19d ago

Evaluation Results

MethodLinks
2026.05
82.21.18289.9143.7
2026.05
82.10.35554.21,082.1
2026.05
70.31.74289.7173.9