| {"./data_dir/eval_vidore/tatdqa_test": {"ndcg_at_1": 0.69927, "ndcg_at_3": 0.78159, "ndcg_at_5": 0.80664, "ndcg_at_10": 0.82137, "ndcg_at_20": 0.82457, "ndcg_at_50": 0.82793, "ndcg_at_100": 0.8296, "map_at_1": 0.69927, "map_at_3": 0.76225, "map_at_5": 0.77616, "map_at_10": 0.78248, "map_at_20": 0.78342, "map_at_50": 0.78395, "map_at_100": 0.7841, "recall_at_1": 0.69927, "recall_at_3": 0.83718, "recall_at_5": 0.89793, "recall_at_10": 0.94228, "recall_at_20": 0.95443, "recall_at_50": 0.97145, "recall_at_100": 0.98177, "precision_at_1": 0.69927, "precision_at_3": 0.27906, "precision_at_5": 0.17959, "precision_at_10": 0.09423, "precision_at_20": 0.04772, "precision_at_50": 0.01943, "precision_at_100": 0.00982, "mrr_at_1": 0.6974483596597812, "mrr_at_3": 0.761543134872418, "mrr_at_5": 0.7746658566221142, "mrr_at_10": 0.7814586587976624, "mrr_at_20": 0.7824191336998677, "mrr_at_50": 0.7829654852280344, "mrr_at_100": 0.7831078050872305, "naucs_at_1_max": 0.25090705114736095, "naucs_at_1_std": -0.2940533354269669, "naucs_at_1_diff1": 0.8007724165936554, "naucs_at_3_max": 0.3012096695305833, "naucs_at_3_std": -0.26907262221168937, "naucs_at_3_diff1": 0.694047297263378, "naucs_at_5_max": 0.38315408734215384, "naucs_at_5_std": -0.11705503964351588, "naucs_at_5_diff1": 0.6525965995076792, "naucs_at_10_max": 0.5023489181540841, "naucs_at_10_std": 0.07188093987240221, "naucs_at_10_diff1": 0.6035975824693318, "naucs_at_20_max": 0.4924317060595406, "naucs_at_20_std": 0.20348352545042372, "naucs_at_20_diff1": 0.564375594315046, "naucs_at_50_max": 0.5257178180099493, "naucs_at_50_std": 0.4387361821625485, "naucs_at_50_diff1": 0.5178667258065841, "naucs_at_100_max": 0.5515774783565448, "naucs_at_100_std": 0.4959428152129429, "naucs_at_100_diff1": 0.48973397557707626}, "./data_dir/eval_vidore/shiftproject_test": {"ndcg_at_1": 0.76, "ndcg_at_3": 0.86202, "ndcg_at_5": 0.87881, "ndcg_at_10": 0.88485, "ndcg_at_20": 0.88485, "ndcg_at_50": 0.88707, "ndcg_at_100": 0.88707, "map_at_1": 0.76, "map_at_3": 0.83833, "map_at_5": 0.84783, "map_at_10": 0.85008, "map_at_20": 0.85008, "map_at_50": 0.85054, "map_at_100": 0.85054, "recall_at_1": 0.76, "recall_at_3": 0.93, "recall_at_5": 0.97, "recall_at_10": 0.99, "recall_at_20": 0.99, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.76, "precision_at_3": 0.31, "precision_at_5": 0.194, "precision_at_10": 0.099, "precision_at_20": 0.0495, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.77, "mrr_at_3": 0.845, "mrr_at_5": 0.8545, "mrr_at_10": 0.85675, "mrr_at_20": 0.85675, "mrr_at_50": 0.8572045454545454, "mrr_at_100": 0.8572045454545454, "naucs_at_1_max": -0.12777970061434624, "naucs_at_1_std": -0.4984857662023017, "naucs_at_1_diff1": 0.6933243921432896, "naucs_at_3_max": 0.5858343337334906, "naucs_at_3_std": -0.004468454048285659, "naucs_at_3_diff1": 0.6212484993997599, "naucs_at_5_max": 0.6101774042950485, "naucs_at_5_std": 0.0704948646125114, "naucs_at_5_diff1": 0.7587924058512269, "naucs_at_10_max": 0.5541549953314738, "naucs_at_10_std": 0.5541549953314738, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 0.5541549953314738, "naucs_at_20_std": 0.5541549953314738, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": NaN, "naucs_at_50_std": NaN, "naucs_at_50_diff1": NaN, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/syntheticDocQA_artificial_intelligence_test": {"ndcg_at_1": 0.99, "ndcg_at_3": 0.995, "ndcg_at_5": 0.995, "ndcg_at_10": 0.995, "ndcg_at_20": 0.995, "ndcg_at_50": 0.995, "ndcg_at_100": 0.995, "map_at_1": 0.99, "map_at_3": 0.99333, "map_at_5": 0.99333, "map_at_10": 0.99333, "map_at_20": 0.99333, "map_at_50": 0.99333, "map_at_100": 0.99333, "recall_at_1": 0.99, "recall_at_3": 1.0, "recall_at_5": 1.0, "recall_at_10": 1.0, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.99, "precision_at_3": 0.33333, "precision_at_5": 0.2, "precision_at_10": 0.1, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.99, "mrr_at_3": 0.9933333333333333, "mrr_at_5": 0.9933333333333333, "mrr_at_10": 0.9933333333333333, "mrr_at_20": 0.9933333333333333, "mrr_at_50": 0.9933333333333333, "mrr_at_100": 0.9933333333333333, "naucs_at_1_max": 0.8692810457516276, "naucs_at_1_std": -0.17133520074696149, "naucs_at_1_diff1": 1.0, "naucs_at_3_max": 1.0, "naucs_at_3_std": 1.0, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 1.0, "naucs_at_5_std": 1.0, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 1.0, "naucs_at_10_std": 1.0, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 1.0, "naucs_at_20_std": 1.0, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": NaN, "naucs_at_50_std": NaN, "naucs_at_50_diff1": NaN, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/syntheticDocQA_government_reports_test": {"ndcg_at_1": 0.95, "ndcg_at_3": 0.98155, "ndcg_at_5": 0.98155, "ndcg_at_10": 0.98155, "ndcg_at_20": 0.98155, "ndcg_at_50": 0.98155, "ndcg_at_100": 0.98155, "map_at_1": 0.95, "map_at_3": 0.975, "map_at_5": 0.975, "map_at_10": 0.975, "map_at_20": 0.975, "map_at_50": 0.975, "map_at_100": 0.975, "recall_at_1": 0.95, "recall_at_3": 1.0, "recall_at_5": 1.0, "recall_at_10": 1.0, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.95, "precision_at_3": 0.33333, "precision_at_5": 0.2, "precision_at_10": 0.1, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.97, "mrr_at_3": 0.985, "mrr_at_5": 0.985, "mrr_at_10": 0.985, "mrr_at_20": 0.985, "mrr_at_50": 0.985, "mrr_at_100": 0.985, "naucs_at_1_max": 0.30924369747899044, "naucs_at_1_std": -0.1360410830999049, "naucs_at_1_diff1": 0.9444444444444438, "naucs_at_3_max": 1.0, "naucs_at_3_std": 1.0, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 1.0, "naucs_at_5_std": 1.0, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 1.0, "naucs_at_10_std": 1.0, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 1.0, "naucs_at_20_std": 1.0, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": NaN, "naucs_at_50_std": NaN, "naucs_at_50_diff1": NaN, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/arxivqa_test_subsampled": {"ndcg_at_1": 0.82, "ndcg_at_3": 0.8565, "ndcg_at_5": 0.8719, "ndcg_at_10": 0.8816, "ndcg_at_20": 0.88566, "ndcg_at_50": 0.89002, "ndcg_at_100": 0.89133, "map_at_1": 0.82, "map_at_3": 0.84767, "map_at_5": 0.85607, "map_at_10": 0.86007, "map_at_20": 0.86119, "map_at_50": 0.86189, "map_at_100": 0.86201, "recall_at_1": 0.82, "recall_at_3": 0.882, "recall_at_5": 0.92, "recall_at_10": 0.95, "recall_at_20": 0.966, "recall_at_50": 0.988, "recall_at_100": 0.996, "precision_at_1": 0.82, "precision_at_3": 0.294, "precision_at_5": 0.184, "precision_at_10": 0.095, "precision_at_20": 0.0483, "precision_at_50": 0.01976, "precision_at_100": 0.00996, "mrr_at_1": 0.818, "mrr_at_3": 0.847, "mrr_at_5": 0.8549, "mrr_at_10": 0.8590880952380953, "mrr_at_20": 0.8602139921739768, "mrr_at_50": 0.8608936438058804, "mrr_at_100": 0.8610132721450209, "naucs_at_1_max": 0.8093071924220723, "naucs_at_1_std": -0.15325476374049982, "naucs_at_1_diff1": 0.9121599295076548, "naucs_at_3_max": 0.8687610005005734, "naucs_at_3_std": -0.10365902888791888, "naucs_at_3_diff1": 0.9062232556637445, "naucs_at_5_max": 0.8757002801120448, "naucs_at_5_std": -0.057691409897293336, "naucs_at_5_diff1": 0.8769257703081252, "naucs_at_10_max": 0.8769374416433241, "naucs_at_10_std": 0.07971988795517813, "naucs_at_10_diff1": 0.8788982259570494, "naucs_at_20_max": 0.8661229197561396, "naucs_at_20_std": 0.04668534080298997, "naucs_at_20_diff1": 0.8863074641621352, "naucs_at_50_max": 0.9074074074074128, "naucs_at_50_std": 0.08784624961095401, "naucs_at_50_diff1": 0.9101307189542546, "naucs_at_100_max": 1.0, "naucs_at_100_std": -0.29878618113917615, "naucs_at_100_diff1": 0.8611111111111168}, "./data_dir/eval_vidore/docvqa_test_subsampled": {"ndcg_at_1": 0.5388, "ndcg_at_3": 0.59518, "ndcg_at_5": 0.61331, "ndcg_at_10": 0.63493, "ndcg_at_20": 0.64571, "ndcg_at_50": 0.66094, "ndcg_at_100": 0.66706, "map_at_1": 0.5388, "map_at_3": 0.58167, "map_at_5": 0.59165, "map_at_10": 0.60065, "map_at_20": 0.60367, "map_at_50": 0.60623, "map_at_100": 0.60677, "recall_at_1": 0.5388, "recall_at_3": 0.63415, "recall_at_5": 0.67849, "recall_at_10": 0.74501, "recall_at_20": 0.78714, "recall_at_50": 0.86253, "recall_at_100": 0.90022, "precision_at_1": 0.5388, "precision_at_3": 0.21138, "precision_at_5": 0.1357, "precision_at_10": 0.0745, "precision_at_20": 0.03936, "precision_at_50": 0.01725, "precision_at_100": 0.009, "mrr_at_1": 0.5388026607538803, "mrr_at_3": 0.581670362158167, "mrr_at_5": 0.591869918699187, "mrr_at_10": 0.6002270087635941, "mrr_at_20": 0.6037472305777859, "mrr_at_50": 0.6061531947736035, "mrr_at_100": 0.6066866021159598, "naucs_at_1_max": 0.6889602472128592, "naucs_at_1_std": -0.1364008689785892, "naucs_at_1_diff1": 0.8930276424851852, "naucs_at_3_max": 0.6662544596828847, "naucs_at_3_std": -0.14933109624836824, "naucs_at_3_diff1": 0.8452905245514425, "naucs_at_5_max": 0.6483204948903895, "naucs_at_5_std": -0.15680650756804887, "naucs_at_5_diff1": 0.8035011466433396, "naucs_at_10_max": 0.6430340754448826, "naucs_at_10_std": -0.08243399370624406, "naucs_at_10_diff1": 0.7638350871217444, "naucs_at_20_max": 0.5999865757022875, "naucs_at_20_std": -0.07009652562261486, "naucs_at_20_diff1": 0.7423981914874048, "naucs_at_50_max": 0.5676739527242458, "naucs_at_50_std": 0.13867806923314227, "naucs_at_50_diff1": 0.7117833895641, "naucs_at_100_max": 0.6025972789679037, "naucs_at_100_std": 0.3436399076261163, "naucs_at_100_diff1": 0.6500925402658911}, "./data_dir/eval_vidore/syntheticDocQA_healthcare_industry_test": {"ndcg_at_1": 0.98, "ndcg_at_3": 0.99131, "ndcg_at_5": 0.99131, "ndcg_at_10": 0.99131, "ndcg_at_20": 0.99131, "ndcg_at_50": 0.99131, "ndcg_at_100": 0.99131, "map_at_1": 0.98, "map_at_3": 0.98833, "map_at_5": 0.98833, "map_at_10": 0.98833, "map_at_20": 0.98833, "map_at_50": 0.98833, "map_at_100": 0.98833, "recall_at_1": 0.98, "recall_at_3": 1.0, "recall_at_5": 1.0, "recall_at_10": 1.0, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.98, "precision_at_3": 0.33333, "precision_at_5": 0.2, "precision_at_10": 0.1, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.98, "mrr_at_3": 0.9883333333333333, "mrr_at_5": 0.9883333333333333, "mrr_at_10": 0.9883333333333333, "mrr_at_20": 0.9883333333333333, "mrr_at_50": 0.9883333333333333, "mrr_at_100": 0.9883333333333333, "naucs_at_1_max": 0.6381886087768457, "naucs_at_1_std": -0.514472455648931, "naucs_at_1_diff1": 1.0, "naucs_at_3_max": 1.0, "naucs_at_3_std": 1.0, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 1.0, "naucs_at_5_std": 1.0, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 1.0, "naucs_at_10_std": 1.0, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 1.0, "naucs_at_20_std": 1.0, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": NaN, "naucs_at_50_std": NaN, "naucs_at_50_diff1": NaN, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/infovqa_test_subsampled": {"ndcg_at_1": 0.86437, "ndcg_at_3": 0.90057, "ndcg_at_5": 0.91233, "ndcg_at_10": 0.91896, "ndcg_at_20": 0.921, "ndcg_at_50": 0.92312, "ndcg_at_100": 0.92379, "map_at_1": 0.86437, "map_at_3": 0.89204, "map_at_5": 0.89862, "map_at_10": 0.9014, "map_at_20": 0.90196, "map_at_50": 0.90235, "map_at_100": 0.90241, "recall_at_1": 0.86437, "recall_at_3": 0.9251, "recall_at_5": 0.95344, "recall_at_10": 0.97368, "recall_at_20": 0.98178, "recall_at_50": 0.9919, "recall_at_100": 0.99595, "precision_at_1": 0.86437, "precision_at_3": 0.30837, "precision_at_5": 0.19069, "precision_at_10": 0.09737, "precision_at_20": 0.04909, "precision_at_50": 0.01984, "precision_at_100": 0.00996, "mrr_at_1": 0.8643724696356275, "mrr_at_3": 0.8927125506072875, "mrr_at_5": 0.8992914979757085, "mrr_at_10": 0.9016748602274918, "mrr_at_20": 0.9022248835445273, "mrr_at_50": 0.9026106475419066, "mrr_at_100": 0.9026754989857776, "naucs_at_1_max": 0.6579798569725905, "naucs_at_1_std": -0.08712140791369336, "naucs_at_1_diff1": 0.9443407469406064, "naucs_at_3_max": 0.7130744631859681, "naucs_at_3_std": -0.06706146302294885, "naucs_at_3_diff1": 0.9080649742289221, "naucs_at_5_max": 0.8099526169769168, "naucs_at_5_std": 0.005146135649768734, "naucs_at_5_diff1": 0.935382109408786, "naucs_at_10_max": 0.861437319577247, "naucs_at_10_std": 0.05225263906280096, "naucs_at_10_diff1": 0.9484969923113604, "naucs_at_20_max": 0.8452245940363352, "naucs_at_20_std": 0.15182628345049073, "naucs_at_20_diff1": 0.9401180393458267, "naucs_at_50_max": 0.7864897480536438, "naucs_at_50_std": 0.6051995005869166, "naucs_at_50_diff1": 0.9305663155605097, "naucs_at_100_max": 0.9346992729675874, "naucs_at_100_std": 0.861132631121016, "naucs_at_100_diff1": 1.0}, "./data_dir/eval_vidore/syntheticDocQA_energy_test": {"ndcg_at_1": 0.93, "ndcg_at_3": 0.94131, "ndcg_at_5": 0.95335, "ndcg_at_10": 0.95335, "ndcg_at_20": 0.95335, "ndcg_at_50": 0.95755, "ndcg_at_100": 0.95755, "map_at_1": 0.93, "map_at_3": 0.93833, "map_at_5": 0.94483, "map_at_10": 0.94483, "map_at_20": 0.94483, "map_at_50": 0.9456, "map_at_100": 0.9456, "recall_at_1": 0.93, "recall_at_3": 0.95, "recall_at_5": 0.98, "recall_at_10": 0.98, "recall_at_20": 0.98, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.93, "precision_at_3": 0.31667, "precision_at_5": 0.196, "precision_at_10": 0.098, "precision_at_20": 0.049, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.93, "mrr_at_3": 0.9383333333333332, "mrr_at_5": 0.9453333333333334, "mrr_at_10": 0.9453333333333334, "mrr_at_20": 0.9458333333333333, "mrr_at_50": 0.9462179487179487, "mrr_at_100": 0.9462179487179487, "naucs_at_1_max": 0.7316259837268236, "naucs_at_1_std": -0.3516073095905031, "naucs_at_1_diff1": 1.0, "naucs_at_3_max": 0.7690009337068138, "naucs_at_3_std": -0.5534080298786143, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 0.861111111111116, "naucs_at_5_std": -1.1517273576097045, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 0.861111111111116, "naucs_at_10_std": -1.1517273576097045, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 0.861111111111116, "naucs_at_20_std": -1.1517273576097045, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": NaN, "naucs_at_50_std": NaN, "naucs_at_50_diff1": NaN, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/tabfquad_test_subsampled": {"ndcg_at_1": 0.81786, "ndcg_at_3": 0.86913, "ndcg_at_5": 0.87927, "ndcg_at_10": 0.8873, "ndcg_at_20": 0.89518, "ndcg_at_50": 0.89824, "ndcg_at_100": 0.89824, "map_at_1": 0.81786, "map_at_3": 0.85714, "map_at_5": 0.86268, "map_at_10": 0.86595, "map_at_20": 0.86798, "map_at_50": 0.86857, "map_at_100": 0.86857, "recall_at_1": 0.81786, "recall_at_3": 0.90357, "recall_at_5": 0.92857, "recall_at_10": 0.95357, "recall_at_20": 0.98571, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.81786, "precision_at_3": 0.30119, "precision_at_5": 0.18571, "precision_at_10": 0.09536, "precision_at_20": 0.04929, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.8142857142857143, "mrr_at_3": 0.8559523809523809, "mrr_at_5": 0.8614880952380952, "mrr_at_10": 0.8648398526077097, "mrr_at_20": 0.8670438372717507, "mrr_at_50": 0.8674632095661231, "mrr_at_100": 0.8674632095661231, "naucs_at_1_max": 0.5880722164938427, "naucs_at_1_std": 0.2677379817513273, "naucs_at_1_diff1": 0.8970448045757864, "naucs_at_3_max": 0.6361137047411553, "naucs_at_3_std": 0.36993809869626837, "naucs_at_3_diff1": 0.843171836635889, "naucs_at_5_max": 0.6008169934640527, "naucs_at_5_std": 0.3147525676937467, "naucs_at_5_diff1": 0.815242763772175, "naucs_at_10_max": 0.5090138619550364, "naucs_at_10_std": 0.26635782518135487, "naucs_at_10_diff1": 0.7559793148028457, "naucs_at_20_max": 0.8068394024276438, "naucs_at_20_std": 0.6692343604108401, "naucs_at_20_diff1": 0.7047152194211078, "naucs_at_50_max": 1.0, "naucs_at_50_std": 1.0, "naucs_at_50_diff1": 1.0, "naucs_at_100_max": 1.0, "naucs_at_100_std": 1.0, "naucs_at_100_diff1": 1.0}} |