List your company
ResearchFormerly WizWand

Mathematical Reasoning on GSM-Infinite 32K

15.4AccuracyFlexPrefill
6.2488.6241113.376Jul 29, 2025
Updated 6mo ago

Evaluation results

MethodLinks
FlexPrefillBackbone Model=Llama-3.1-8B-Inst2025.0715.4
MInferenceBackbone Model=Qwen2.5-7B-Inst2025.0715.1
XAttentionBackbone Model=Qwen2.5-7B-Inst2025.0714.3
Ours + XAttentionBackbone Model=Qwen2.5-7B-Inst2025.0714.1
FlexPrefillBackbone Model=Qwen2.5-7B-Inst2025.0713.4
Ours + FlexPrefillBackbone Model=Qwen2.5-7B-Inst2025.0713.4
Ours + MInferenceBackbone Model=Qwen2.5-7B-Inst2025.0713.3
TriangleMixBackbone Model=Qwen2.5-7B-Inst2025.0712.9
Qwen2.5-7B-InstBackbone Model=Qwen2.5-7B-Inst2025.0712.8
Ours + XAttentionBackbone Model=Llama-3.1-8B-Inst2025.0712
XAttentionBackbone Model=Llama-3.1-8B-Inst2025.0710.6
FlexPrefillBackbone Model=Llama-3-8B-Inst-262k2025.0710.2
Llama-3.1-8B-InstBackbone Model=Llama-3.1-8B-Inst2025.079.6
TriangleMixBackbone Model=Llama-3.1-8B-Inst2025.079.6
Ours + MInferenceBackbone Model=Llama-3.1-8B-Inst2025.078.9
Ours + FlexPrefillBackbone Model=Llama-3.1-8B-Inst2025.078.9
XAttentionBackbone Model=Llama-3-8B-Inst-262k2025.078.2
MInferenceBackbone Model=Llama-3.1-8B-Inst2025.078.1
TriangleMixBackbone Model=Llama-3-8B-Inst-262k2025.078.1
Llama-3-8B-Inst-262kBackbone Model=Llama-3-8B-Inst-262k2025.077.9
MInferenceBackbone Model=Llama-3-8B-Inst-262k2025.077.5
Ours + XAttentionBackbone Model=Llama-3-8B-Inst-262k2025.077.4
Ours + MInferenceBackbone Model=Llama-3-8B-Inst-262k2025.076.6
Ours + FlexPrefillBackbone Model=Llama-3-8B-Inst-262k2025.076.6