Multi-Hop Question Answering on HotpotQA (test val)
47.1Exact Match (EM)DeepControl
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DeepControlBackbone Model=Qwen2.5-7B2026.02 | 47.1 | 47.9 | — | |
| (ParaSet+SetCE)+CE5Base=Contriever, Latency (sec/query)=0.1962026.07 | 42.9 | — | 55.3 | |
| (ParaSet+SetCE)+CE5Base=Qwen3, Latency (sec/query)=0.2412026.07 | 41 | — | 52.9 | |
| ParaSet+SetCEBase=Contriever, Latency (sec/query)=0.1162026.07 | 40.4 | — | 52.2 | |
| DeepControlBackbone Model=Qwen2.5-3B2026.02 | 40.2 | 41.1 | — | |
| CE10Base=Contriever, Latency (sec/query)=0.0672026.07 | 40.1 | — | 51.6 | |
| CE10Base=Qwen3, Latency (sec/query)=0.1142026.07 | 39.7 | — | 51.5 | |
| SetCEBase=Qwen3, Latency (sec/query)=0.5852026.07 | 39.6 | — | 50.8 | |
| SetCEBase=Contriever, Latency (sec/query)=0.5552026.07 | 39.1 | — | 50.6 | |
| CE5Base=Contriever, Latency (sec/query)=0.0672026.07 | 39 | — | 50.6 | |
| ListCEBase=Contriever, Latency (sec/query)=0.5552026.07 | 38.7 | — | 49.7 | |
| CE5Base=Qwen3, Latency (sec/query)=0.1142026.07 | 38.4 | — | 49.4 | |
| ListCEBase=Qwen3, Latency (sec/query)=0.5852026.07 | 38 | — | 48.6 | |
| BiBase=Qwen3, Latency (sec/query)=0.0732026.07 | 37 | — | 47.3 | |
| ParaSet+SetCEBase=Qwen3, Latency (sec/query)=0.1642026.07 | 36.9 | — | 48 | |
| BiBase=Contriever, Latency (sec/query)=0.0322026.07 | 35 | — | 45.9 | |
| Direct InferenceBackbone Model=Qwen2.5-7B2026.02 | 18.3 | 18.1 | — | |
| Direct InferenceBackbone Model=Qwen2.5-3B2026.02 | 14.9 | 13.4 | — |