List your company

Instruction Following on Alpaca-Eval (test)

66.85Length-Controlled WinrateNo Defense
35.80643.865551.92559.9845Feb 2, 2026

Evaluation results

MethodLinks
No DefenseBackbone=Vicuna2026.0266.8566.831.648051,108
NGSDBackbone=Vicuna2026.0265.1764.181.698051,015
No DefenseBackbone=Llama3-8b2026.0263.4964.631.698051,616
NGSDBackbone=Llama3-8b2026.0256.7953.731.768051,366
SSDBackbone=Vicuna2026.0255.8757.891.748051,155
SSDBackbone=Llama3-8b2026.023737.021.78051,378