Mathematical Reasoning on GSM-Infinite
22.9Accuracy (8K)TriangleMix
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TriangleMixBase Model=Qwen2.5-7B-Inst2025.07 | 22.9 | 14.9 | 12.9 | 16.9 | |
| Qwen2.5-7B-InstBase Model=Qwen2.5-7B-Inst2025.07 | 21 | 14.6 | 12.8 | 16.1 | |
| StreamingMixBase Model=Qwen2.5-7B-Inst2025.07 | 20.3 | 14.4 | 14.6 | 16.4 | |
| TriangleBase Model=Qwen2.5-7B-Inst2025.07 | 14.6 | 9.9 | 10.8 | 11.7 | |
| StreamingBase Model=Qwen2.5-7B-Inst2025.07 | 12.3 | 8.4 | 7.6 | 9.4 | |
| StreamingBase Model=Llama-3.1-8B-Inst2025.07 | 8.3 | 4.6 | 7.4 | 6.8 | |
| TriangleMixBase Model=Llama-3.1-8B-Inst2025.07 | 7 | 5.8 | 9.6 | 7.4 | |
| Llama-3.1-8B-InstBase Model=Llama-3.1-8B-Inst2025.07 | 6.9 | 6.4 | 9.6 | 7.6 | |
| DuoAttentionBase Model=Llama-3.1-8B-Inst2025.07 | 6.7 | 5.6 | 9.1 | 7.1 | |
| TriangleBase Model=Llama-3.1-8B-Inst2025.07 | 6.4 | 4.9 | 9.3 | 6.8 | |
| StreamingMixBase Model=Llama-3.1-8B-Inst2025.07 | 6.3 | 4.7 | 7.8 | 6.3 | |
| Llama-3-8B-Inst-262kBase Model=Llama-3-8B-Inst-262k2025.07 | 3.7 | 6.4 | 7.9 | 6 | |
| TriangleMixBase Model=Llama-3-8B-Inst-262k2025.07 | 3.4 | 6.7 | 8.1 | 6 | |
| DuoAttentionBase Model=Llama-3-8B-Inst-262k2025.07 | 3.1 | 4.9 | 5.5 | 4.5 | |
| StreamingMixBase Model=Llama-3-8B-Inst-262k2025.07 | 2.7 | 3.6 | 4.6 | 3.6 | |
| StreamingBase Model=Llama-3-8B-Inst-262k2025.07 | 2 | 3.3 | 1.5 | 2.3 | |
| TriangleBase Model=Llama-3-8B-Inst-262k2025.07 | 1.9 | 6.2 | 6.2 | 4.7 |