Spaces:
Sleeping
Sleeping
| { | |
| "dataset": "hover", | |
| "questions": 300, | |
| "candidate_count_total": 3000, | |
| "candidate_pool": "10 HotpotQA distractor passages, only HoVer claims with all support titles present", | |
| "gold_support_count_distribution": { | |
| "2": 152, | |
| "3": 144, | |
| "4": 4 | |
| }, | |
| "gold_available_fraction": 100.0, | |
| "gold_available_count": 300, | |
| "graph_link_fraction": 94.33, | |
| "oracle_complete_coverage_pct": { | |
| "2": 50.67, | |
| "3": 98.67, | |
| "5": 100.0, | |
| "8": 100.0 | |
| }, | |
| "complete_coverage_pct": { | |
| "dense_cosine": { | |
| "2": 26.33, | |
| "3": 47.33, | |
| "5": 75.67, | |
| "8": 92.0 | |
| }, | |
| "crossencoder": { | |
| "2": 33.0, | |
| "3": 57.0, | |
| "5": 82.0, | |
| "8": 92.33 | |
| }, | |
| "dense_graph": { | |
| "2": 32.33, | |
| "3": 73.33, | |
| "5": 89.33, | |
| "8": 97.33 | |
| }, | |
| "crossencoder_graph": { | |
| "2": 38.33, | |
| "3": 72.67, | |
| "5": 92.33, | |
| "8": 98.67 | |
| } | |
| }, | |
| "crossencoder_graph_minus_crossencoder_at3_pp": 15.67, | |
| "ci95_pp": [ | |
| 9.99, | |
| 22.0 | |
| ], | |
| "paired_wins_losses_ties_at3": { | |
| "wins": 70, | |
| "losses": 23, | |
| "ties": 207 | |
| }, | |
| "timing_seconds": { | |
| "embedding_seconds": 45.51700891600012, | |
| "prep_seconds": 45.92742176499996, | |
| "candidate_pool_size": 10, | |
| "eligibility": "HoVer dev claims with HotpotQA dev passages available and all annotated support titles present; conditional subset", | |
| "crossencoder_scoring_seconds": 84.113, | |
| "crossencoder_scoring_ms_per_question": 280.38, | |
| "crossencoder_forward_seconds": 83.542, | |
| "graph_selection_total_s": 0.11, | |
| "graph_selection_ms_per_question": 0.37 | |
| }, | |
| "models": { | |
| "dense": "user uploaded all-MiniLM-L6-v2", | |
| "crossencoder": "user uploaded cross-encoder/ms-marco-MiniLM-L6-v2" | |
| }, | |
| "protocol": "MiniLM embeddings at max 128 WordPiece for dense. Cross-encoder truncates paired query+passage at 256 WordPiece. Both score exact same candidate pools. Fixed evidence graph .4 minmax relevance + .5 title in already-selected passage + .2 query-title score. No TF-IDF anywhere in candidate construction or scoring. Gold used solely for pool audit and evaluation." | |
| } |