Grade-school math on GSM8K
92.4AccuracyCLSR
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CLSRLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 92.4 | — | — | — | 198 | |
| Raw CoTLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 92.2 | — | — | — | 433 | |
| SoTLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 91.8 | — | — | — | 205 | |
| CLSRLLM=QWEN3-32B2026.06 | 91.6 | — | — | — | 91 | |
| Raw CoTLLM=QWEN3-32B2026.06 | 91.3 | — | — | — | 216 | |
| CLSRLLM=QWEN3-8B2026.06 | 91.2 | — | — | — | 89 | |
| Raw CoTLLM=QWEN3-8B2026.06 | 90.9 | — | — | — | 243 | |
| SoTLLM=QWEN3-32B2026.06 | 90.5 | — | — | — | 112 | |
| CCoTLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 90.3 | — | — | — | 220 | |
| SoTLLM=QWEN3-8B2026.06 | 90.3 | — | — | — | 107 | |
| CCoTLLM=QWEN3-8B2026.06 | 89.2 | — | — | — | 125 | |
| CCoTLLM=QWEN3-32B2026.06 | 88.2 | — | — | — | 94 | |
| CoDLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 86.5 | — | — | — | 204 | |
| CoDLLM=QWEN3-8B2026.06 | 85.4 | — | — | — | 100 | |
| Raw CoTLLM=LLaMA3-8B2026.06 | 84.9 | — | — | — | 124 | |
| CLSRLLM=LLaMA3-8B2026.06 | 84.3 | — | — | — | 52 | |
| CCoTLLM=LLaMA3-8B2026.06 | 83.6 | — | — | — | 65 | |
| SoTLLM=LLaMA3-8B2026.06 | 82.3 | — | — | — | 55 | |
| CoDLLM=QWEN3-32B2026.06 | 82.1 | — | — | — | 105 | |
| CoDLLM=LLaMA3-8B2026.06 | 80.5 | — | — | — | 51 | |
| MoE-Sieve (Qwen1.5-MoE-A2.7B)Condition=Hot (25%)2026.03 | 59.2 | 0.2 | 0.77 | — | — | |
| Qwen1.5-MoE-A2.7BCondition=Full LoRA2026.03 | 59 | — | — | — | — | |
| OLMoE-1B-7BCondition=Full LoRA2026.03 | 30.4 | — | — | — | — | |
| MoE-Sieve (OLMoE-1B-7B)Condition=Hot (25%)2026.03 | 30.4 | 0.08 | 1.45 | — | — |