List your company
ResearchFormerly WizWand

Mathematical Reasoning on GSM-Infinite 16K

16.2AccuracyOurs + MInference
4.8647.80710.7513.693Jul 29, 2025
Updated 6mo ago

Evaluation results

MethodLinks
Ours + MInferenceBackbone Model=Qwen2.5-7B-Inst2025.0716.2
XAttentionBackbone Model=Qwen2.5-7B-Inst2025.0715.9
MInferenceBackbone Model=Qwen2.5-7B-Inst2025.0715.5
TriangleMixBackbone Model=Qwen2.5-7B-Inst2025.0714.9
Qwen2.5-7B-InstBackbone Model=Qwen2.5-7B-Inst2025.0714.6
FlexPrefillBackbone Model=Qwen2.5-7B-Inst2025.0714.5
Ours + XAttentionBackbone Model=Qwen2.5-7B-Inst2025.0714.4
Ours + FlexPrefillBackbone Model=Qwen2.5-7B-Inst2025.0713.8
FlexPrefillBackbone Model=Llama-3.1-8B-Inst2025.0711.2
FlexPrefillBackbone Model=Llama-3-8B-Inst-262k2025.078.7
Ours + XAttentionBackbone Model=Llama-3.1-8B-Inst2025.076.8
XAttentionBackbone Model=Llama-3-8B-Inst-262k2025.076.8
XAttentionBackbone Model=Llama-3.1-8B-Inst2025.076.7
TriangleMixBackbone Model=Llama-3-8B-Inst-262k2025.076.7
Ours + XAttentionBackbone Model=Llama-3-8B-Inst-262k2025.076.6
Llama-3.1-8B-InstBackbone Model=Llama-3.1-8B-Inst2025.076.4
Llama-3-8B-Inst-262kBackbone Model=Llama-3-8B-Inst-262k2025.076.4
MInferenceBackbone Model=Llama-3-8B-Inst-262k2025.076.3
Ours + MInferenceBackbone Model=Llama-3-8B-Inst-262k2025.076.1
Ours + FlexPrefillBackbone Model=Llama-3-8B-Inst-262k2025.076.1
TriangleMixBackbone Model=Llama-3.1-8B-Inst2025.075.8
MInferenceBackbone Model=Llama-3.1-8B-Inst2025.075.8
Ours + MInferenceBackbone Model=Llama-3.1-8B-Inst2025.075.3
Ours + FlexPrefillBackbone Model=Llama-3.1-8B-Inst2025.075.3