List your company
ResearchFormerly WizWand

Mathematical Reasoning on GPQA Diamond (test)

67.34Accuracy@5LARK
9.214424.304739.39554.4853May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
LARKBackbone=Qwen-2.5-7B, Selection Budget (B)=32026.0567.34
LARKBackbone=Qwen-2.5-1.5B, Selection Budget (B)=32026.0566.67
LARKBackbone=Qwen-2.5-7B, Selection Budget (B)=12026.0565.99
LARKBackbone=Qwen-2.5-1.5B, Selection Budget (B)=12026.0563.13
RandomBackbone=Qwen-2.5-7B, Selection Budget (B)=32026.0559.76
RandomBackbone=Qwen-2.5-7B, Selection Budget (B)=12026.0556.9
RSRminBackbone=Qwen-2.5-7B, Selection Budget (B)=32026.0556.73
Local NaturalnessmaxBackbone=Qwen-2.5-7B, Selection Budget (B)=32026.0556.23
RandomBackbone=Qwen-2.5-1.5B, Selection Budget (B)=32026.0555.89
GRAPEmaxBackbone=Qwen-2.5-7B, Selection Budget (B)=32026.0555.05
LARKBackbone=Llama-3.2-3B, Selection Budget (B)=32026.0554.55
RandomBackbone=Qwen-2.5-1.5B, Selection Budget (B)=12026.0554.38
Rule-based QualitymaxBackbone=Qwen-2.5-7B, Selection Budget (B)=32026.0553.54
LLM-judged QualitymaxBackbone=Qwen-2.5-7B, Selection Budget (B)=32026.0553.03
RSRminBackbone=Qwen-2.5-7B, Selection Budget (B)=12026.0552.19
Local NaturalnessmaxBackbone=Qwen-2.5-7B, Selection Budget (B)=12026.0551.18
LLM-judged QualitymaxBackbone=Qwen-2.5-7B, Selection Budget (B)=12026.0550.34
GRAPEmaxBackbone=Qwen-2.5-7B, Selection Budget (B)=12026.0549.33
LARKBackbone=Llama-3.2-3B, Selection Budget (B)=12026.0548.32
Rule-based QualitymaxBackbone=Qwen-2.5-7B, Selection Budget (B)=12026.0548.15
RSRminBackbone=Qwen-2.5-1.5B, Selection Budget (B)=12026.0547.14
Local NaturalnessmaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=32026.0546.8
GRAPEmaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=32026.0546.63
Rule-based QualitymaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=32026.0546.3
LLM-judged QualitymaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=32026.0546.3
RSRminBackbone=Qwen-2.5-1.5B, Selection Budget (B)=32026.0546.3
Local NaturalnessmaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=12026.0545.62
RandomBackbone=Llama-3.2-3B, Selection Budget (B)=32026.0544.61
Token LengthmaxBackbone=Qwen-2.5-7B, Selection Budget (B)=32026.0543.77
RSRminBackbone=Llama-3.2-3B, Selection Budget (B)=32026.0543.77
LLM-judged QualitymaxBackbone=Llama-3.2-3B, Selection Budget (B)=32026.0543.43
LLM-judged QualitymaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=12026.0542.93
GRAPEmaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=12026.0542.93
Local NaturalnessmaxBackbone=Llama-3.2-3B, Selection Budget (B)=32026.0542.76
RSRminBackbone=Llama-3.2-3B, Selection Budget (B)=12026.0542.09
Token LengthmaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=32026.0542.09
LLM-judged QualitymaxBackbone=Llama-3.2-3B, Selection Budget (B)=12026.0541.58
Rule-based QualitymaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=12026.0541.25
Token LengthmaxBackbone=Qwen-2.5-7B, Selection Budget (B)=12026.0541.08
Local NaturalnessmaxBackbone=Llama-3.2-3B, Selection Budget (B)=12026.0541.08
RandomBackbone=Llama-3.2-3B, Selection Budget (B)=12026.0540.4
BaseModelBackbone=Qwen-2.5-7B, Selection Budget (B)=12026.0539.56
BaseModelBackbone=Qwen-2.5-7B, Selection Budget (B)=32026.0539.56
Token LengthmaxBackbone=Llama-3.2-3B, Selection Budget (B)=32026.0538.72
Token LengthmaxBackbone=Llama-3.2-3B, Selection Budget (B)=12026.0538.22
Token LengthmaxBackbone=Qwen-2.5-1.5B, Selection Budget (B)=12026.0534.51
BaseModelBackbone=Qwen-2.5-1.5B, Selection Budget (B)=12026.0530.64
BaseModelBackbone=Qwen-2.5-1.5B, Selection Budget (B)=32026.0530.64
Rule-based QualitymaxBackbone=Llama-3.2-3B, Selection Budget (B)=32026.0523.57
GRAPEmaxBackbone=Llama-3.2-3B, Selection Budget (B)=32026.0521.55
Rule-based QualitymaxBackbone=Llama-3.2-3B, Selection Budget (B)=12026.0514.48
GRAPEmaxBackbone=Llama-3.2-3B, Selection Budget (B)=12026.0514.31
BaseModelBackbone=Llama-3.2-3B, Selection Budget (B)=12026.0511.45
BaseModelBackbone=Llama-3.2-3B, Selection Budget (B)=32026.0511.45