List your company

Mathematical Reasoning on AIME 26 (pass@1 accuracy, standard error)

72.3Pass@1 AccuracyVASE-ATTNV
59.71662.98366.2569.517Jun 2, 2026

Evaluation results

MethodLinks
VASE-ATTNVBase Model=Qwen3-14B, Cache Compression=~ 4x2026.0672.36.5
VASE-DKVBase Model=Qwen3-14B, Cache Compression=~ 4x2026.0668.86.9
CurDKVBase Model=Qwen3-14B, Cache Compression=~ 4x2026.0661.37.3
R-KVBase Model=Qwen3-14B, Cache Compression=~ 4x2026.0660.27.4