Multi-hop Question Answering on InfoQA Synthetic Multi-hop QA 1-hop
100Average F1Direct
Evaluation results
| Method | Links | |
|---|---|---|
| DirectContext Length=0.5k, Backbone=Qwen3-14B2025.09 | 100 | |
| Chain-of-Thought (CoT)Context Length=0.5k, Backbone=Qwen3-14B2025.09 | 100 | |
| Self-Refine (S-R)Context Length=0.5k, Backbone=Qwen3-14B2025.09 | 100 | |
| Self Consistency (S-C)Context Length=0.5k, Backbone=Qwen3-14B2025.09 | 100 | |
| ReActContext Length=0.5k, Backbone=Qwen3-14B2025.09 | 100 | |
| Plan-and-Solve (P&S)Context Length=0.5k, Backbone=Qwen3-14B2025.09 | 100 | |
| Self-Ask (S-A)Context Length=0.5k, Backbone=Qwen3-14B2025.09 | 100 | |
| InfoQAContext Length=0.5k, Backbone=Qwen3-14B2025.09 | 100 | |
| InfoQA w/o Pruning Past Reasoning Trace (P.)Context Length=0.5k, Backbone=Qwen3-14B2025.09 | 100 | |
| DirectContext Length=1k, Backbone=Qwen3-14B2025.09 | 100 | |
| Chain-of-Thought (CoT)Context Length=1k, Backbone=Qwen3-14B2025.09 | 100 | |
| Self-Refine (S-R)Context Length=1k, Backbone=Qwen3-14B2025.09 | 100 | |
| Self Consistency (S-C)Context Length=1k, Backbone=Qwen3-14B2025.09 | 100 | |
| Plan-and-Solve (P&S)Context Length=1k, Backbone=Qwen3-14B2025.09 | 100 | |
| Self-Ask (S-A)Context Length=1k, Backbone=Qwen3-14B2025.09 | 100 | |
| InfoQAContext Length=1k, Backbone=Qwen3-14B2025.09 | 100 | |
| InfoQA w/o Pruning Past Reasoning Trace (P.)Context Length=1k, Backbone=Qwen3-14B2025.09 | 100 | |
| ReActContext Length=1k, Backbone=Qwen3-14B2025.09 | 99 | |
| InfoQAContext Length=Average, Backbone=Qwen3-14B2025.09 | 99 | |
| DirectContext Length=Average, Backbone=Qwen3-14B2025.09 | 97 | |
| InfoQAContext Length=10k, Backbone=Qwen3-14B2025.09 | 96 | |
| DirectContext Length=10k, Backbone=Qwen3-14B2025.09 | 91 | |
| InfoQA w/o Capacity-Aware Task Decomposition (D.)Context Length=0.5k, Backbone=Qwen3-14B2025.09 | 87 | |
| InfoQA w/o Capacity-Aware Task Decomposition (D.)Context Length=1k, Backbone=Qwen3-14B2025.09 | 78 |