List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME25 (Accuracy (avg@32))

15.6Accuracy (avg@32)LESS
3.3286.5149.712.886Nov 30, 2025
Updated 8mo ago

Evaluation results

MethodLinks
LESSBackbone=Qwen2.5-Math-7B2025.1115.6
Forking TokensBackbone=Qwen2.5-Math-7B2025.1114.2
KL-CovBackbone=Qwen2.5-Math-7B2025.1114.1
GRPOBackbone=Qwen2.5-Math-7B2025.1113.8
LESSBackbone=Qwen2.5-Base-7B2025.1113.1
LESSBackbone=Qwen2.5-Math-1.5B2025.1112.4
Forking TokensBackbone=Qwen2.5-Base-7B2025.1112.3
GRPOBackbone=Qwen2.5-Base-7B2025.1111.9
KL-CovBackbone=Qwen2.5-Base-7B2025.1110.8
KL-CovBackbone=Qwen2.5-Math-1.5B2025.119.2
Base LLMBackbone=Qwen2.5-Math-7B2025.118.9
Forking TokensBackbone=Qwen2.5-Math-1.5B2025.117
GRPOBackbone=Qwen2.5-Math-1.5B2025.116.6
Base LLMBackbone=Qwen2.5-Base-7B2025.115.2
Base LLMBackbone=Qwen2.5-Math-1.5B2025.113.8