Multi-hop Question Answering on HOTPOT-QA (ACC, TKN)
68.8AccuracyCLSR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CLSRLLM=QWEN3-32B2026.06 | 68.8 | 55 | |
| Raw CoTLLM=QWEN3-32B2026.06 | 68.4 | 165 | |
| Raw CoTLLM=QWEN3-8B2026.06 | 66.4 | 134 | |
| CLSRLLM=QWEN3-8B2026.06 | 66.4 | 48 | |
| SoTLLM=QWEN3-32B2026.06 | 66.2 | 72 | |
| SoTLLM=QWEN3-8B2026.06 | 65.8 | 63 | |
| CCoTLLM=QWEN3-32B2026.06 | 65.8 | 81 | |
| CCoTLLM=QWEN3-8B2026.06 | 65.2 | 70 | |
| CoDLLM=QWEN3-32B2026.06 | 64.6 | 78 | |
| CoDLLM=QWEN3-8B2026.06 | 63.2 | 58 | |
| Raw CoTLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 55.8 | 213 | |
| SoTLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 54.8 | 59 | |
| CLSRLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 54.5 | 43 | |
| CCoTLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 53 | 65 | |
| CoDLLM=DEEPSEEK-R1-0528-QWEN3-8B2026.06 | 51.2 | 56 | |
| SoTLLM=LLaMA3-8B2026.06 | 45.8 | 21 | |
| CLSRLLM=LLaMA3-8B2026.06 | 45.5 | 14 | |
| Raw CoTLLM=LLaMA3-8B2026.06 | 45.1 | 32 | |
| CCoTLLM=LLaMA3-8B2026.06 | 44.3 | 28 | |
| CoDLLM=LLaMA3-8B2026.06 | 40.2 | 24 |