Multi-Hop Question Answering on Synthetic Multi-Hop Benchmark 3-hop
92Avg F1InfoQA
Evaluation results
| Method | Links | |
|---|---|---|
| InfoQAContext Length=0.5k, Backbone=Qwen3-8B2025.09 | 92 | |
| Chain-of-Thought (CoT)Context Length=4k, Backbone=Qwen3-8B2025.09 | 60 | |
| DirectContext Length=0.5k, Backbone=Qwen3-8B2025.09 | 50 |