Multi-Hop Question Answering on 2Wiki (EM, FBERT)
47.91EMCoT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CoTSLM=Qwen3-235B2026.04 | 47.91 | 76.81 | |
| Dynamic Collaboration FrameworkSLM=Qwen3-8B2026.04 | 47.09 | 77.35 | |
| Direct InferSLM=Qwen3-235B2026.04 | 41.57 | 72.4 | |
| Dynamic Collaboration FrameworkSLM=Qwen3-4B2026.04 | 40.52 | 74.33 | |
| Dynamic Collaboration FrameworkSLM=Llama-3.1-8B2026.04 | 39.27 | 73.23 | |
| Dynamic Collaboration FrameworkSLM=Llama-3.2-3B2026.04 | 38.33 | 72.3 | |
| CoTSLM=Qwen3-8B2026.04 | 33.51 | 68.65 | |
| CoTSLM=Qwen3-4B2026.04 | 32.2 | 68.26 | |
| CoTSLM=Llama-3.1-8B2026.04 | 22.44 | 58.69 | |
| CoTSLM=Llama-3.2-3B2026.04 | 21.82 | 58.63 |