Multi-Hop Question Answering on HotpotQA (EM, FBERT)
30.81Exact Match (EM)CoT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CoTSLM=Qwen3-235B2026.04 | 30.81 | 69.26 | |
| Dynamic Collaboration FrameworkSLM=Qwen3-8B2026.04 | 29.66 | 69.21 | |
| Dynamic Collaboration FrameworkSLM=Llama-3.1-8B2026.04 | 25.54 | 67.06 | |
| Dynamic Collaboration FrameworkSLM=Qwen3-4B2026.04 | 25.51 | 67.3 | |
| Direct InferSLM=Qwen3-235B2026.04 | 25 | 65.35 | |
| Dynamic Collaboration FrameworkSLM=Llama-3.2-3B2026.04 | 22.77 | 64.96 | |
| CoTSLM=Qwen3-8B2026.04 | 15.15 | 59.59 | |
| CoTSLM=Llama-3.1-8B2026.04 | 14.82 | 55.31 | |
| CoTSLM=Qwen3-4B2026.04 | 10.98 | 57.41 | |
| CoTSLM=Llama-3.2-3B2026.04 | 9.42 | 51.36 |