Spaces:
Sleeping
Sleeping
| dataset,n_questions,budget_k,candidate_pool,dense_cosine_pct,crossencoder_pct,dense_plus_graph_pct,crossencoder_plus_graph_pct,crossencoder_graph_gain_pp,crossencoder_cpu_scoring_ms_per_question,graph_selection_ms_per_question,graph_timing_scope,crossencoder_inference_source | |
| HotpotQA,600,2,native 10 Wikipedia distractor candidates,33.5,45.83,61.67,67.17,21.34,272.18,0.406,feature extraction + 4 rank orders,previously measured original pretrained forward pass; cached scores reused for unified rerun | |
| HotpotQA,600,3,native 10 Wikipedia distractor candidates,50.67,60.83,74.83,80.5,19.67,272.18,0.406,feature extraction + 4 rank orders,previously measured original pretrained forward pass; cached scores reused for unified rerun | |
| HotpotQA,600,5,native 10 Wikipedia distractor candidates,68.0,74.67,84.83,89.83,15.16,272.18,0.406,feature extraction + 4 rank orders,previously measured original pretrained forward pass; cached scores reused for unified rerun | |
| HotpotQA,600,8,native 10 Wikipedia distractor candidates,85.67,86.83,94.17,95.5,8.67,272.18,0.406,feature extraction + 4 rank orders,previously measured original pretrained forward pass; cached scores reused for unified rerun | |
| 2WikiMultiHopQA,600,2,native 10 Wikipedia candidates,24.83,33.83,61.5,62.83,29.0,254.92,0.322,feature extraction + 4 rank orders,previously measured original pretrained forward pass; cached scores reused for unified rerun | |
| 2WikiMultiHopQA,600,3,native 10 Wikipedia candidates,34.17,44.5,68.33,70.17,25.67,254.92,0.322,feature extraction + 4 rank orders,previously measured original pretrained forward pass; cached scores reused for unified rerun | |
| 2WikiMultiHopQA,600,5,native 10 Wikipedia candidates,52.33,58.67,93.0,93.17,34.5,254.92,0.322,feature extraction + 4 rank orders,previously measured original pretrained forward pass; cached scores reused for unified rerun | |
| 2WikiMultiHopQA,600,8,native 10 Wikipedia candidates,81.5,79.33,98.33,98.17,18.84,254.92,0.322,feature extraction + 4 rank orders,previously measured original pretrained forward pass; cached scores reused for unified rerun | |
| MuSiQue,300,2,"native ~20 candidates, all gold included",9.67,11.33,16.33,17.67,6.34,538.67,0.662,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| MuSiQue,300,3,"native ~20 candidates, all gold included",16.0,19.33,25.0,27.0,7.67,538.67,0.662,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| MuSiQue,300,5,"native ~20 candidates, all gold included",31.0,30.67,43.67,42.33,11.66,538.67,0.662,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| MuSiQue,300,8,"native ~20 candidates, all gold included",44.0,44.33,58.67,57.67,13.34,538.67,0.662,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| HoVer,300,2,10 matched HotpotQA candidates; eligible subset only,26.33,33.0,32.33,38.33,5.33,280.38,0.37,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| HoVer,300,3,10 matched HotpotQA candidates; eligible subset only,47.33,57.0,73.33,72.67,15.67,280.38,0.37,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| HoVer,300,5,10 matched HotpotQA candidates; eligible subset only,75.67,82.0,89.33,92.33,10.33,280.38,0.37,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| HoVer,300,8,10 matched HotpotQA candidates; eligible subset only,92.0,92.33,97.33,98.67,6.34,280.38,0.37,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| MultiHop-RAG,300,2,"MiniLM top-20 from full 609-article corpus, no gold insertion",6.0,7.0,6.0,7.67,0.67,635.04,7.57,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| MultiHop-RAG,300,3,"MiniLM top-20 from full 609-article corpus, no gold insertion",11.67,13.67,11.67,13.67,0.0,635.04,7.57,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| MultiHop-RAG,300,5,"MiniLM top-20 from full 609-article corpus, no gold insertion",21.67,23.0,22.33,26.0,3.0,635.04,7.57,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| MultiHop-RAG,300,8,"MiniLM top-20 from full 609-article corpus, no gold insertion",31.67,36.67,32.33,36.67,0.0,635.04,7.57,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| QASC,300,2,"MiniLM top-20 from 6,878 fact-sentence pool, no gold insertion",3.33,7.33,3.33,7.33,0.0,69.12,0.34,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| QASC,300,3,"MiniLM top-20 from 6,878 fact-sentence pool, no gold insertion",5.33,13.0,5.33,13.0,0.0,69.12,0.34,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| QASC,300,5,"MiniLM top-20 from 6,878 fact-sentence pool, no gold insertion",10.33,17.67,10.33,17.67,0.0,69.12,0.34,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |
| QASC,300,8,"MiniLM top-20 from 6,878 fact-sentence pool, no gold insertion",16.0,24.33,16.0,24.33,0.0,69.12,0.34,reported feature extraction + evaluation/selection operation,previous pretrained neural inference | |