vidore-benchmark computed results

results.json

@@ -1 +1 @@
-{"validation_set": {"ndcg_at_1": 0.842, "ndcg_at_3": 0.88607, "ndcg_at_5": 0.89166, "ndcg_at_10": 0.89809, "ndcg_at_20": 0.90151, "ndcg_at_50": 0.90388, "ndcg_at_100": 0.90649, "map_at_1": 0.842, "map_at_3": 0.87567, "map_at_5": 0.87867, "map_at_10": 0.8813, "map_at_20": 0.88218, "map_at_50": 0.88255, "map_at_100": 0.88279, "recall_at_1": 0.842, "recall_at_3": 0.916, "recall_at_5": 0.93, "recall_at_10": 0.95, "recall_at_20": 0.964, "recall_at_50": 0.976, "recall_at_100": 0.992, "precision_at_1": 0.842, "precision_at_3": 0.30533, "precision_at_5": 0.186, "precision_at_10": 0.095, "precision_at_20": 0.0482, "precision_at_50": 0.01952, "precision_at_100": 0.00992, "mrr_at_1": 0.846, "mrr_at_3": 0.878, "mrr_at_5": 0.8817999999999999, "mrr_at_10": 0.884202380952381, "mrr_at_20": 0.8851428457507404, "mrr_at_50": 0.8859005862850743, "mrr_at_100": 0.8859630560187304, "naucs_at_1_max": 0.14980322536592605, "naucs_at_1_std": 0.13490546127202088, "naucs_at_1_diff1": 0.9576655203664858, "naucs_at_3_max": -0.1255502200880325, "naucs_at_3_std": 0.4357854252812249, "naucs_at_3_diff1": 0.9442554799697651, "naucs_at_5_max": -0.18676804054955595, "naucs_at_5_std": 0.5007869814592494, "naucs_at_5_diff1": 0.9691876750700267, "naucs_at_10_max": -0.24179271708683783, "naucs_at_10_std": 0.5684407096171821, "naucs_at_10_diff1": 0.9732026143790856, "naucs_at_20_max": -0.3797333748314225, "naucs_at_20_std": 0.6160908807967637, "naucs_at_20_diff1": 0.985475671750177, "naucs_at_50_max": -0.709617180205425, "naucs_at_50_std": 0.7341269841269842, "naucs_at_50_diff1": 1.0, "naucs_at_100_max": -0.21755368814195147, "naucs_at_100_std": 0.9346405228758211, "naucs_at_100_diff1": 1.0}, "syntheticDocQA_energy": {"ndcg_at_1": 0.95, "ndcg_at_3": 0.96131, "ndcg_at_5": 0.96131, "ndcg_at_10": 0.96788, "ndcg_at_20": 0.9702, "ndcg_at_50": 0.9702, "ndcg_at_100": 0.9702, "map_at_1": 0.95, "map_at_3": 0.95833, "map_at_5": 0.95833, "map_at_10": 0.96111, "map_at_20": 0.96164, "map_at_50": 0.96164, "map_at_100": 0.96164, "recall_at_1": 0.95, "recall_at_3": 0.97, "recall_at_5": 0.97, "recall_at_10": 0.99, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.95, "precision_at_3": 0.32333, "precision_at_5": 0.194, "precision_at_10": 0.099, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.96, "mrr_at_3": 0.9633333333333333, "mrr_at_5": 0.9653333333333333, "mrr_at_10": 0.9664444444444444, "mrr_at_20": 0.9671587301587301, "mrr_at_50": 0.9671587301587301, "mrr_at_100": 0.9671587301587301, "naucs_at_1_max": 0.683099906629318, "naucs_at_1_std": -0.6536881419234336, "naucs_at_1_diff1": 1.0, "naucs_at_3_max": 0.664021164021167, "naucs_at_3_std": -1.21708683473389, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 0.6640211640211615, "naucs_at_5_std": -1.2170868347338937, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 0.8692810457516413, "naucs_at_10_std": -1.7399626517273863, "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}, "syntheticDocQA_healthcare_industry": {"ndcg_at_1": 0.96, "ndcg_at_3": 0.98524, "ndcg_at_5": 0.98524, "ndcg_at_10": 0.98524, "ndcg_at_20": 0.98524, "ndcg_at_50": 0.98524, "ndcg_at_100": 0.98524, "map_at_1": 0.96, "map_at_3": 0.98, "map_at_5": 0.98, "map_at_10": 0.98, "map_at_20": 0.98, "map_at_50": 0.98, "map_at_100": 0.98, "recall_at_1": 0.96, "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": 0.96, "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": 0.96, "mrr_at_3": 0.98, "mrr_at_5": 0.98, "mrr_at_10": 0.98, "mrr_at_20": 0.98, "mrr_at_50": 0.98, "mrr_at_100": 0.98, "naucs_at_1_max": 0.8284313725490191, "naucs_at_1_std": -0.2200046685340767, "naucs_at_1_diff1": 0.96732026143791, "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}, "syntheticDocQA_artificial_intelligence_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": 0.8692810457516332, "naucs_at_1_std": -0.5088702147525628, "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}, "syntheticDocQA_government_reports": {"ndcg_at_1": 0.9, "ndcg_at_3": 0.95417, "ndcg_at_5": 0.95417, "ndcg_at_10": 0.95718, "ndcg_at_20": 0.95718, "ndcg_at_50": 0.95718, "ndcg_at_100": 0.95718, "map_at_1": 0.9, "map_at_3": 0.94167, "map_at_5": 0.94167, "map_at_10": 0.94278, "map_at_20": 0.94278, "map_at_50": 0.94278, "map_at_100": 0.94278, "recall_at_1": 0.9, "recall_at_3": 0.99, "recall_at_5": 0.99, "recall_at_10": 1.0, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.9, "precision_at_3": 0.33, "precision_at_5": 0.198, "precision_at_10": 0.1, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.9, "mrr_at_3": 0.9433333333333332, "mrr_at_5": 0.9433333333333332, "mrr_at_10": 0.9444444444444444, "mrr_at_20": 0.9444444444444444, "mrr_at_50": 0.9444444444444444, "mrr_at_100": 0.9444444444444444, "naucs_at_1_max": 0.6324463118580764, "naucs_at_1_std": 0.3140989729225021, "naucs_at_1_diff1": 0.9607843137254906, "naucs_at_3_max": 1.0, "naucs_at_3_std": 0.7222222222222157, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 1.0, "naucs_at_5_std": 0.7222222222222276, "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}, "infovqa_subsampled": {"ndcg_at_1": 0.884, "ndcg_at_3": 0.91245, "ndcg_at_5": 0.91968, "ndcg_at_10": 0.92228, "ndcg_at_20": 0.92482, "ndcg_at_50": 0.92647, "ndcg_at_100": 0.92777, "map_at_1": 0.884, "map_at_3": 0.90567, "map_at_5": 0.90957, "map_at_10": 0.91066, "map_at_20": 0.91135, "map_at_50": 0.91165, "map_at_100": 0.91176, "recall_at_1": 0.884, "recall_at_3": 0.932, "recall_at_5": 0.95, "recall_at_10": 0.958, "recall_at_20": 0.968, "recall_at_50": 0.976, "recall_at_100": 0.984, "precision_at_1": 0.884, "precision_at_3": 0.31067, "precision_at_5": 0.19, "precision_at_10": 0.0958, "precision_at_20": 0.0484, "precision_at_50": 0.01952, "precision_at_100": 0.00984, "mrr_at_1": 0.882, "mrr_at_3": 0.9043333333333333, "mrr_at_5": 0.9081333333333333, "mrr_at_10": 0.9092444444444444, "mrr_at_20": 0.9099755336167101, "mrr_at_50": 0.9102621834483599, "mrr_at_100": 0.9103823466185231, "naucs_at_1_max": 0.7065231646939233, "naucs_at_1_std": 0.05565756254303531, "naucs_at_1_diff1": 0.940899701629562, "naucs_at_3_max": 0.8456774866809487, "naucs_at_3_std": 0.30691492283187727, "naucs_at_3_diff1": 0.9336793540945822, "naucs_at_5_max": 0.941176470588235, "naucs_at_5_std": 0.46224089635854493, "naucs_at_5_diff1": 0.9313725490196085, "naucs_at_10_max": 0.942421413009648, "naucs_at_10_std": 0.4962651727357598, "naucs_at_10_diff1": 0.9245253657018353, "naucs_at_20_max": 0.9581290849673211, "naucs_at_20_std": 0.6626400560224076, "naucs_at_20_diff1": 0.917279411764704, "naucs_at_50_max": 0.978213507625274, "naucs_at_50_std": 0.8516184251478357, "naucs_at_50_diff1": 0.911492374727667, "naucs_at_100_max": 0.9673202614379109, "naucs_at_100_std": 0.8740079365079331, "naucs_at_100_diff1": 0.8672385620914999}, "docvqa_subsampled": {"ndcg_at_1": 0.482, "ndcg_at_3": 0.53638, "ndcg_at_5": 0.5566, "ndcg_at_10": 0.58035, "ndcg_at_20": 0.59013, "ndcg_at_50": 0.60114, "ndcg_at_100": 0.60886, "map_at_1": 0.482, "map_at_3": 0.52333, "map_at_5": 0.53433, "map_at_10": 0.54403, "map_at_20": 0.5468, "map_at_50": 0.54853, "map_at_100": 0.54919, "recall_at_1": 0.482, "recall_at_3": 0.574, "recall_at_5": 0.624, "recall_at_10": 0.698, "recall_at_20": 0.736, "recall_at_50": 0.792, "recall_at_100": 0.84, "precision_at_1": 0.482, "precision_at_3": 0.19133, "precision_at_5": 0.1248, "precision_at_10": 0.0698, "precision_at_20": 0.0368, "precision_at_50": 0.01584, "precision_at_100": 0.0084, "mrr_at_1": 0.476, "mrr_at_3": 0.5206666666666666, "mrr_at_5": 0.5325666666666667, "mrr_at_10": 0.5408849206349206, "mrr_at_20": 0.5443294747306358, "mrr_at_50": 0.5456530086612842, "mrr_at_100": 0.5466288691712617, "naucs_at_1_max": 0.11546090350805194, "naucs_at_1_std": 0.678213639746092, "naucs_at_1_diff1": 0.9150091230002886, "naucs_at_3_max": -0.006613409596074608, "naucs_at_3_std": 0.7932113091033843, "naucs_at_3_diff1": 0.8907490104390393, "naucs_at_5_max": -0.07645913914509621, "naucs_at_5_std": 0.8597608598631361, "naucs_at_5_diff1": 0.8752417435287114, "naucs_at_10_max": -0.18047119686299265, "naucs_at_10_std": 0.8799692900476072, "naucs_at_10_diff1": 0.8267973035513766, "naucs_at_20_max": -0.2529875372440978, "naucs_at_20_std": 0.888227661551276, "naucs_at_20_diff1": 0.8310159228526568, "naucs_at_50_max": -0.38866116247068644, "naucs_at_50_std": 0.8807246545341788, "naucs_at_50_diff1": 0.825517956470337, "naucs_at_100_max": -0.465617329405992, "naucs_at_100_std": 0.9135125184094256, "naucs_at_100_diff1": 0.8196305841924404}, "arxivqa_subsampled": {"ndcg_at_1": 0.842, "ndcg_at_3": 0.88255, "ndcg_at_5": 0.88935, "ndcg_at_10": 0.89974, "ndcg_at_20": 0.90386, "ndcg_at_50": 0.90791, "ndcg_at_100": 0.90858, "map_at_1": 0.842, "map_at_3": 0.873, "map_at_5": 0.8769, "map_at_10": 0.88121, "map_at_20": 0.88239, "map_at_50": 0.88307, "map_at_100": 0.88314, "recall_at_1": 0.842, "recall_at_3": 0.91, "recall_at_5": 0.926, "recall_at_10": 0.958, "recall_at_20": 0.974, "recall_at_50": 0.994, "recall_at_100": 0.998, "precision_at_1": 0.842, "precision_at_3": 0.30333, "precision_at_5": 0.1852, "precision_at_10": 0.0958, "precision_at_20": 0.0487, "precision_at_50": 0.01988, "precision_at_100": 0.00998, "mrr_at_1": 0.846, "mrr_at_3": 0.8753333333333334, "mrr_at_5": 0.8791333333333333, "mrr_at_10": 0.883215873015873, "mrr_at_20": 0.8845011145287461, "mrr_at_50": 0.8851056968145472, "mrr_at_100": 0.8851712618465302, "naucs_at_1_max": 0.762746899402848, "naucs_at_1_std": -0.09687891842232711, "naucs_at_1_diff1": 0.93131510012539, "naucs_at_3_max": 0.7958190683680891, "naucs_at_3_std": -0.08053740014524166, "naucs_at_3_diff1": 0.9046581595601199, "naucs_at_5_max": 0.8209932621697326, "naucs_at_5_std": -0.06355515179044569, "naucs_at_5_diff1": 0.913632119514472, "naucs_at_10_max": 0.9291938997821336, "naucs_at_10_std": 0.06940553999377147, "naucs_at_10_diff1": 0.9618736383442255, "naucs_at_20_max": 0.9384112619406718, "naucs_at_20_std": 0.2580621992386671, "naucs_at_20_diff1": 0.9899446958270457, "naucs_at_50_max": 0.9564270152505304, "naucs_at_50_std": -0.061935885465295104, "naucs_at_50_diff1": 1.0, "naucs_at_100_max": 1.0, "naucs_at_100_std": 1.0, "naucs_at_100_diff1": 1.0}, "tabfquad_subsampled": {"ndcg_at_1": 0.85714, "ndcg_at_3": 0.90616, "ndcg_at_5": 0.91046, "ndcg_at_10": 0.91828, "ndcg_at_20": 0.92089, "ndcg_at_50": 0.92319, "ndcg_at_100": 0.92382, "map_at_1": 0.85714, "map_at_3": 0.89464, "map_at_5": 0.89696, "map_at_10": 0.90003, "map_at_20": 0.90069, "map_at_50": 0.90113, "map_at_100": 0.9012, "recall_at_1": 0.85714, "recall_at_3": 0.93929, "recall_at_5": 0.95, "recall_at_10": 0.975, "recall_at_20": 0.98571, "recall_at_50": 0.99643, "recall_at_100": 1.0, "precision_at_1": 0.85714, "precision_at_3": 0.3131, "precision_at_5": 0.19, "precision_at_10": 0.0975, "precision_at_20": 0.04929, "precision_at_50": 0.01993, "precision_at_100": 0.01, "mrr_at_1": 0.8571428571428571, "mrr_at_3": 0.8946428571428572, "mrr_at_5": 0.89625, "mrr_at_10": 0.8999489795918367, "mrr_at_20": 0.9006087018140589, "mrr_at_50": 0.9010484076109075, "mrr_at_100": 0.9011170889295889, "naucs_at_1_max": 0.5042274052478143, "naucs_at_1_std": 0.15470116618076035, "naucs_at_1_diff1": 0.8651239067055385, "naucs_at_3_max": 0.6153127917833795, "naucs_at_3_std": 0.4349426044927801, "naucs_at_3_diff1": 0.7869500741473071, "naucs_at_5_max": 0.5725623582766437, "naucs_at_5_std": 0.4422102174203035, "naucs_at_5_diff1": 0.7964852607709757, "naucs_at_10_max": 0.4158329998666156, "naucs_at_10_std": 0.32713085234094164, "naucs_at_10_diff1": 0.6116446578631451, "naucs_at_20_max": 0.24964985994398026, "naucs_at_20_std": 0.21288515406162958, "naucs_at_20_diff1": 0.6785714285714314, "naucs_at_50_max": 0.8692810457515607, "naucs_at_50_std": 0.8692810457515607, "naucs_at_50_diff1": 0.7222222222221809, "naucs_at_100_max": 1.0, "naucs_at_100_std": 1.0, "naucs_at_100_diff1": 1.0}, "tatdqa": {"ndcg_at_1": 0.69934, "ndcg_at_3": 0.79233, "ndcg_at_5": 0.81055, "ndcg_at_10": 0.82355, "ndcg_at_20": 0.82931, "ndcg_at_50": 0.83237, "ndcg_at_100": 0.83391, "map_at_1": 0.69934, "map_at_3": 0.76959, "map_at_5": 0.77979, "map_at_10": 0.78535, "map_at_20": 0.78692, "map_at_50": 0.78745, "map_at_100": 0.78758, "recall_at_1": 0.69934, "recall_at_3": 0.85809, "recall_at_5": 0.90198, "recall_at_10": 0.94107, "recall_at_20": 0.96392, "recall_at_50": 0.97895, "recall_at_100": 0.98857, "precision_at_1": 0.69934, "precision_at_3": 0.28603, "precision_at_5": 0.1804, "precision_at_10": 0.09411, "precision_at_20": 0.0482, "precision_at_50": 0.01958, "precision_at_100": 0.00989, "mrr_at_1": 0.6945279615153337, "mrr_at_3": 0.7672880336740829, "mrr_at_5": 0.7785628382441371, "mrr_at_10": 0.7839064990598364, "mrr_at_20": 0.7854265248894812, "mrr_at_50": 0.7860095907963216, "mrr_at_100": 0.7861276188247445, "naucs_at_1_max": 0.22893544807358054, "naucs_at_1_std": -0.10645073148522476, "naucs_at_1_diff1": 0.820977721729132, "naucs_at_3_max": 0.36359547295706673, "naucs_at_3_std": 0.04589921689064907, "naucs_at_3_diff1": 0.7721105105285038, "naucs_at_5_max": 0.38687174851887957, "naucs_at_5_std": 0.13613442362048297, "naucs_at_5_diff1": 0.7514418550740282, "naucs_at_10_max": 0.515265589904788, "naucs_at_10_std": 0.41696526710652737, "naucs_at_10_diff1": 0.7441091180365171, "naucs_at_20_max": 0.5145946804670753, "naucs_at_20_std": 0.46286538260906074, "naucs_at_20_diff1": 0.7308007486507647, "naucs_at_50_max": 0.6663489691190478, "naucs_at_50_std": 0.6527939062776273, "naucs_at_50_diff1": 0.7597957531590336, "naucs_at_100_max": 0.6806290990559558, "naucs_at_100_std": 0.7236588802071027, "naucs_at_100_diff1": 0.7904400313210926}, "shift_project": {"ndcg_at_1": 0.8, "ndcg_at_3": 0.8844, "ndcg_at_5": 0.89258, "ndcg_at_10": 0.89915, "ndcg_at_20": 0.90194, "ndcg_at_50": 0.90398, "ndcg_at_100": 0.90398, "map_at_1": 0.8, "map_at_3": 0.865, "map_at_5": 0.8695, "map_at_10": 0.87228, "map_at_20": 0.87319, "map_at_50": 0.87353, "map_at_100": 0.87353, "recall_at_1": 0.8, "recall_at_3": 0.94, "recall_at_5": 0.96, "recall_at_10": 0.98, "recall_at_20": 0.99, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.8, "precision_at_3": 0.31333, "precision_at_5": 0.192, "precision_at_10": 0.098, "precision_at_20": 0.0495, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.79, "mrr_at_3": 0.8633333333333333, "mrr_at_5": 0.8658333333333332, "mrr_at_10": 0.8686111111111111, "mrr_at_20": 0.8695202020202021, "mrr_at_50": 0.869920202020202, "mrr_at_100": 0.869920202020202, "naucs_at_1_max": 0.023051948051947945, "naucs_at_1_std": -0.4089660339660347, "naucs_at_1_diff1": 0.7242257742257738, "naucs_at_3_max": 0.18285091814503548, "naucs_at_3_std": -0.27653283535636064, "naucs_at_3_diff1": 0.6640211640211647, "naucs_at_5_max": -0.1562791783379993, "naucs_at_5_std": -0.6954948646125065, "naucs_at_5_diff1": 0.9346405228758139, "naucs_at_10_max": -0.43534080298785716, "naucs_at_10_std": -0.43534080298785716, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": -1.7399626517273863, "naucs_at_20_std": -1.7399626517273863, "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}}
+{"../colpali/data_dir/eval_vidore/tatdqa_test": {"ndcg_at_1": 0.69259, "ndcg_at_3": 0.79041, "ndcg_at_5": 0.80995, "ndcg_at_10": 0.82297, "ndcg_at_20": 0.82756, "ndcg_at_50": 0.83099, "ndcg_at_100": 0.83246, "map_at_1": 0.69259, "map_at_3": 0.7661, "map_at_5": 0.7771, "map_at_10": 0.78262, "map_at_20": 0.78387, "map_at_50": 0.78444, "map_at_100": 0.78457, "recall_at_1": 0.69259, "recall_at_3": 0.86087, "recall_at_5": 0.90765, "recall_at_10": 0.94714, "recall_at_20": 0.96537, "recall_at_50": 0.98238, "recall_at_100": 0.99149, "precision_at_1": 0.69259, "precision_at_3": 0.28696, "precision_at_5": 0.18153, "precision_at_10": 0.09471, "precision_at_20": 0.04827, "precision_at_50": 0.01965, "precision_at_100": 0.00991, "mrr_at_1": 0.6938031591737546, "mrr_at_3": 0.7682260024301336, "mrr_at_5": 0.7792223572296476, "mrr_at_10": 0.784609394588131, "mrr_at_20": 0.7859477041844983, "mrr_at_50": 0.7864718344490268, "mrr_at_100": 0.7865677750728132, "naucs_at_1_max": 0.1784395098159134, "naucs_at_1_std": 0.05860002752594418, "naucs_at_1_diff1": 0.8058592501770308, "naucs_at_3_max": 0.22799093261457798, "naucs_at_3_std": 0.16649813864435306, "naucs_at_3_diff1": 0.7450317457466235, "naucs_at_5_max": 0.2553901794657091, "naucs_at_5_std": 0.23663082281817802, "naucs_at_5_diff1": 0.7194106841420028, "naucs_at_10_max": 0.19611099032314872, "naucs_at_10_std": 0.21572189203593398, "naucs_at_10_diff1": 0.7283887824534098, "naucs_at_20_max": 0.27585284137046934, "naucs_at_20_std": 0.24072881821977038, "naucs_at_20_diff1": 0.74424158271136, "naucs_at_50_max": 0.28621399304289236, "naucs_at_50_std": 0.3763275538695369, "naucs_at_50_diff1": 0.7209026229080127, "naucs_at_100_max": 0.4321788328506148, "naucs_at_100_std": 0.4735592017300837, "naucs_at_100_diff1": 0.793431320847917}, "../colpali/data_dir/eval_vidore/shiftproject_test": {"ndcg_at_1": 0.72, "ndcg_at_3": 0.83964, "ndcg_at_5": 0.85643, "ndcg_at_10": 0.85944, "ndcg_at_20": 0.86444, "ndcg_at_50": 0.86646, "ndcg_at_100": 0.86646, "map_at_1": 0.72, "map_at_3": 0.81167, "map_at_5": 0.82117, "map_at_10": 0.82228, "map_at_20": 0.82362, "map_at_50": 0.82395, "map_at_100": 0.82395, "recall_at_1": 0.72, "recall_at_3": 0.92, "recall_at_5": 0.96, "recall_at_10": 0.97, "recall_at_20": 0.99, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.72, "precision_at_3": 0.30667, "precision_at_5": 0.192, "precision_at_10": 0.097, "precision_at_20": 0.0495, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.73, "mrr_at_3": 0.8166666666666668, "mrr_at_5": 0.8266666666666667, "mrr_at_10": 0.8277777777777777, "mrr_at_20": 0.8292063492063493, "mrr_at_50": 0.8295396825396826, "mrr_at_100": 0.8295396825396826, "naucs_at_1_max": 0.15647893772893828, "naucs_at_1_std": -0.19957646520146544, "naucs_at_1_diff1": 0.860653235653236, "naucs_at_3_max": 0.23541083099906737, "naucs_at_3_std": -0.15750466853408215, "naucs_at_3_diff1": 0.8176353874883276, "naucs_at_5_max": -0.19712885154061158, "naucs_at_5_std": -0.34010270774976264, "naucs_at_5_diff1": 0.8978758169934612, "naucs_at_10_max": -0.20572673513850018, "naucs_at_10_std": -0.3963585434173691, "naucs_at_10_diff1": 0.9074074074074053, "naucs_at_20_max": -1.7399626517273863, "naucs_at_20_std": -1.7399626517273863, "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}, "../colpali/data_dir/eval_vidore/syntheticDocQA_artificial_intelligence_test": {"ndcg_at_1": 0.98, "ndcg_at_3": 0.99262, "ndcg_at_5": 0.99262, "ndcg_at_10": 0.99262, "ndcg_at_20": 0.99262, "ndcg_at_50": 0.99262, "ndcg_at_100": 0.99262, "map_at_1": 0.98, "map_at_3": 0.99, "map_at_5": 0.99, "map_at_10": 0.99, "map_at_20": 0.99, "map_at_50": 0.99, "map_at_100": 0.99, "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": 0.99, "mrr_at_5": 0.99, "mrr_at_10": 0.99, "mrr_at_20": 0.99, "mrr_at_50": 0.99, "mrr_at_100": 0.99, "naucs_at_1_max": 0.795751633986929, "naucs_at_1_std": 0.6136788048552733, "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}, "../colpali/data_dir/eval_vidore/syntheticDocQA_government_reports_test": {"ndcg_at_1": 0.88, "ndcg_at_3": 0.94678, "ndcg_at_5": 0.94678, "ndcg_at_10": 0.94994, "ndcg_at_20": 0.94994, "ndcg_at_50": 0.94994, "ndcg_at_100": 0.94994, "map_at_1": 0.88, "map_at_3": 0.93167, "map_at_5": 0.93167, "map_at_10": 0.93292, "map_at_20": 0.93292, "map_at_50": 0.93292, "map_at_100": 0.93292, "recall_at_1": 0.88, "recall_at_3": 0.99, "recall_at_5": 0.99, "recall_at_10": 1.0, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.88, "precision_at_3": 0.33, "precision_at_5": 0.198, "precision_at_10": 0.1, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.89, "mrr_at_3": 0.9366666666666668, "mrr_at_5": 0.9366666666666668, "mrr_at_10": 0.9380952380952381, "mrr_at_20": 0.9380952380952381, "mrr_at_50": 0.9380952380952381, "mrr_at_100": 0.9380952380952381, "naucs_at_1_max": 0.4139755011135863, "naucs_at_1_std": 0.025214762965320433, "naucs_at_1_diff1": 0.9415367483296211, "naucs_at_3_max": 1.0, "naucs_at_3_std": 0.7222222222222157, "naucs_at_3_diff1": 0.8692810457516356, "naucs_at_5_max": 1.0, "naucs_at_5_std": 0.7222222222222276, "naucs_at_5_diff1": 0.8692810457516413, "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}, "../colpali/data_dir/eval_vidore/arxivqa_test_subsampled": {"ndcg_at_1": 0.786, "ndcg_at_3": 0.84638, "ndcg_at_5": 0.85696, "ndcg_at_10": 0.86769, "ndcg_at_20": 0.87313, "ndcg_at_50": 0.87632, "ndcg_at_100": 0.87766, "map_at_1": 0.786, "map_at_3": 0.83133, "map_at_5": 0.83713, "map_at_10": 0.8414, "map_at_20": 0.84282, "map_at_50": 0.84334, "map_at_100": 0.84347, "recall_at_1": 0.786, "recall_at_3": 0.89, "recall_at_5": 0.916, "recall_at_10": 0.95, "recall_at_20": 0.972, "recall_at_50": 0.988, "recall_at_100": 0.996, "precision_at_1": 0.786, "precision_at_3": 0.29667, "precision_at_5": 0.1832, "precision_at_10": 0.095, "precision_at_20": 0.0486, "precision_at_50": 0.01976, "precision_at_100": 0.00996, "mrr_at_1": 0.79, "mrr_at_3": 0.8336666666666667, "mrr_at_5": 0.8380666666666667, "mrr_at_10": 0.8428373015873015, "mrr_at_20": 0.8444688522262052, "mrr_at_50": 0.8449797466991158, "mrr_at_100": 0.8451291335696891, "naucs_at_1_max": 0.7084097446322541, "naucs_at_1_std": 0.263068109122443, "naucs_at_1_diff1": 0.9287064810997545, "naucs_at_3_max": 0.7413881968903021, "naucs_at_3_std": 0.4373765140451864, "naucs_at_3_diff1": 0.8924061506743405, "naucs_at_5_max": 0.7259348183717956, "naucs_at_5_std": 0.46819839046729866, "naucs_at_5_diff1": 0.8677804455115391, "naucs_at_10_max": 0.8406162464986007, "naucs_at_10_std": 0.6695424836601321, "naucs_at_10_diff1": 0.8545284780578905, "naucs_at_20_max": 0.8409363745498202, "naucs_at_20_std": 0.8039549152994563, "naucs_at_20_diff1": 0.8514405762304914, "naucs_at_50_max": 0.863834422657961, "naucs_at_50_std": 0.7394179894179909, "naucs_at_50_diff1": 0.8856209150326784, "naucs_at_100_max": 0.8611111111111168, "naucs_at_100_std": 1.0, "naucs_at_100_diff1": 0.9346405228758466}, "../colpali/data_dir/eval_vidore/docvqa_test_subsampled": {"ndcg_at_1": 0.49446, "ndcg_at_3": 0.56507, "ndcg_at_5": 0.583, "ndcg_at_10": 0.59997, "ndcg_at_20": 0.61131, "ndcg_at_50": 0.62086, "ndcg_at_100": 0.62986, "map_at_1": 0.49446, "map_at_3": 0.54804, "map_at_5": 0.5578, "map_at_10": 0.5651, "map_at_20": 0.56827, "map_at_50": 0.56976, "map_at_100": 0.57055, "recall_at_1": 0.49446, "recall_at_3": 0.61419, "recall_at_5": 0.65854, "recall_at_10": 0.70953, "recall_at_20": 0.75388, "recall_at_50": 0.80266, "recall_at_100": 0.85809, "precision_at_1": 0.49446, "precision_at_3": 0.20473, "precision_at_5": 0.13171, "precision_at_10": 0.07095, "precision_at_20": 0.03769, "precision_at_50": 0.01605, "precision_at_100": 0.00858, "mrr_at_1": 0.5077605321507761, "mrr_at_3": 0.5539541759053954, "mrr_at_5": 0.5640428677014042, "mrr_at_10": 0.5700929152148665, "mrr_at_20": 0.5731150426232905, "mrr_at_50": 0.5748682542132003, "mrr_at_100": 0.575487754001377, "naucs_at_1_max": -0.1606923114398016, "naucs_at_1_std": 0.7856291230396629, "naucs_at_1_diff1": 0.8942494117954668, "naucs_at_3_max": -0.3463973185357886, "naucs_at_3_std": 0.8519388680212596, "naucs_at_3_diff1": 0.8293610487503834, "naucs_at_5_max": -0.4659091345071243, "naucs_at_5_std": 0.8744182624949535, "naucs_at_5_diff1": 0.8143175138411842, "naucs_at_10_max": -0.5280895802868046, "naucs_at_10_std": 0.893712207848969, "naucs_at_10_diff1": 0.7799562131421047, "naucs_at_20_max": -0.6403408069604808, "naucs_at_20_std": 0.9149410793311721, "naucs_at_20_diff1": 0.7626617810805572, "naucs_at_50_max": -0.7843420633333069, "naucs_at_50_std": 0.9067555931613859, "naucs_at_50_diff1": 0.7214352021905747, "naucs_at_100_max": -0.8399926782810305, "naucs_at_100_std": 0.8848389472066827, "naucs_at_100_diff1": 0.6986609787107426}, "../colpali/data_dir/eval_vidore/syntheticDocQA_healthcare_industry_test": {"ndcg_at_1": 0.95, "ndcg_at_3": 0.96893, "ndcg_at_5": 0.9728, "ndcg_at_10": 0.97613, "ndcg_at_20": 0.97613, "ndcg_at_50": 0.97613, "ndcg_at_100": 0.97613, "map_at_1": 0.95, "map_at_3": 0.965, "map_at_5": 0.967, "map_at_10": 0.96843, "map_at_20": 0.96843, "map_at_50": 0.96843, "map_at_100": 0.96843, "recall_at_1": 0.95, "recall_at_3": 0.98, "recall_at_5": 0.99, "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": 0.32667, "precision_at_5": 0.198, "precision_at_10": 0.1, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.96, "mrr_at_3": 0.97, "mrr_at_5": 0.9725, "mrr_at_10": 0.9741666666666667, "mrr_at_20": 0.9741666666666667, "mrr_at_50": 0.9741666666666667, "mrr_at_100": 0.9741666666666667, "naucs_at_1_max": 0.8627450980392154, "naucs_at_1_std": -0.014192343604110945, "naucs_at_1_diff1": 0.9477124183006508, "naucs_at_3_max": 1.0, "naucs_at_3_std": -0.2987861811391324, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 1.0, "naucs_at_5_std": 0.5541549953314738, "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}, "../colpali/data_dir/eval_vidore/infovqa_test_subsampled": {"ndcg_at_1": 0.87652, "ndcg_at_3": 0.90866, "ndcg_at_5": 0.91372, "ndcg_at_10": 0.91974, "ndcg_at_20": 0.92348, "ndcg_at_50": 0.92595, "ndcg_at_100": 0.9266, "map_at_1": 0.87652, "map_at_3": 0.90148, "map_at_5": 0.90432, "map_at_10": 0.90688, "map_at_20": 0.90799, "map_at_50": 0.90841, "map_at_100": 0.90847, "recall_at_1": 0.87652, "recall_at_3": 0.92915, "recall_at_5": 0.9413, "recall_at_10": 0.95951, "recall_at_20": 0.97368, "recall_at_50": 0.98583, "recall_at_100": 0.98988, "precision_at_1": 0.87652, "precision_at_3": 0.30972, "precision_at_5": 0.18826, "precision_at_10": 0.09595, "precision_at_20": 0.04868, "precision_at_50": 0.01972, "precision_at_100": 0.0099, "mrr_at_1": 0.8765182186234818, "mrr_at_3": 0.9008097165991903, "mrr_at_5": 0.9047570850202429, "mrr_at_10": 0.9073171711329606, "mrr_at_20": 0.9081808466334543, "mrr_at_50": 0.9085825830549277, "mrr_at_100": 0.9086325878746693, "naucs_at_1_max": 0.5904536974572032, "naucs_at_1_std": 0.23046908082706039, "naucs_at_1_diff1": 0.9449631270983153, "naucs_at_3_max": 0.6145285656654232, "naucs_at_3_std": 0.25341007845415703, "naucs_at_3_diff1": 0.9233582467230783, "naucs_at_5_max": 0.6460728176810125, "naucs_at_5_std": 0.31734383640906494, "naucs_at_5_diff1": 0.925515325916096, "naucs_at_10_max": 0.634378549036623, "naucs_at_10_std": 0.4274637190725395, "naucs_at_10_diff1": 0.9181175133912983, "naucs_at_20_max": 0.7810286723433896, "naucs_at_20_std": 0.6042139536876944, "naucs_at_20_diff1": 0.9598149372108383, "naucs_at_50_max": 0.7923359134481396, "naucs_at_50_std": 0.9363278834338821, "naucs_at_50_diff1": 1.0, "naucs_at_100_max": 0.7353905696403603, "naucs_at_100_std": 0.9108590368074757, "naucs_at_100_diff1": 1.0}, "../colpali/data_dir/eval_vidore/syntheticDocQA_energy_test": {"ndcg_at_1": 0.95, "ndcg_at_3": 0.96, "ndcg_at_5": 0.96, "ndcg_at_10": 0.96301, "ndcg_at_20": 0.96799, "ndcg_at_50": 0.96799, "ndcg_at_100": 0.96799, "map_at_1": 0.95, "map_at_3": 0.95667, "map_at_5": 0.95667, "map_at_10": 0.95778, "map_at_20": 0.95911, "map_at_50": 0.95911, "map_at_100": 0.95911, "recall_at_1": 0.95, "recall_at_3": 0.97, "recall_at_5": 0.97, "recall_at_10": 0.98, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.95, "precision_at_3": 0.32333, "precision_at_5": 0.194, "precision_at_10": 0.098, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.95, "mrr_at_3": 0.9566666666666667, "mrr_at_5": 0.9566666666666667, "mrr_at_10": 0.9579166666666667, "mrr_at_20": 0.9593055555555555, "mrr_at_50": 0.9593055555555555, "mrr_at_100": 0.9593055555555555, "naucs_at_1_max": 0.5351073762838467, "naucs_at_1_std": -0.08281979458449801, "naucs_at_1_diff1": 1.0, "naucs_at_3_max": 0.4663865546218464, "naucs_at_3_std": -0.29831932773109476, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 0.466386554621846, "naucs_at_5_std": -0.2983193277310949, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 0.6381886087768404, "naucs_at_10_std": 0.42250233426704475, "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}, "../colpali/data_dir/eval_vidore/tabfquad_test_subsampled": {"ndcg_at_1": 0.83214, "ndcg_at_3": 0.88474, "ndcg_at_5": 0.89641, "ndcg_at_10": 0.90316, "ndcg_at_20": 0.90667, "ndcg_at_50": 0.90877, "ndcg_at_100": 0.90938, "map_at_1": 0.83214, "map_at_3": 0.87202, "map_at_5": 0.87845, "map_at_10": 0.88113, "map_at_20": 0.88203, "map_at_50": 0.88236, "map_at_100": 0.88242, "recall_at_1": 0.83214, "recall_at_3": 0.92143, "recall_at_5": 0.95, "recall_at_10": 0.97143, "recall_at_20": 0.98571, "recall_at_50": 0.99643, "recall_at_100": 1.0, "precision_at_1": 0.83214, "precision_at_3": 0.30714, "precision_at_5": 0.19, "precision_at_10": 0.09714, "precision_at_20": 0.04929, "precision_at_50": 0.01993, "precision_at_100": 0.01, "mrr_at_1": 0.8357142857142857, "mrr_at_3": 0.8726190476190476, "mrr_at_5": 0.8801190476190476, "mrr_at_10": 0.8828713151927438, "mrr_at_20": 0.8837846805388823, "mrr_at_50": 0.8841111920065087, "mrr_at_100": 0.8841727683611885, "naucs_at_1_max": 0.494567611217149, "naucs_at_1_std": 0.24208737965895424, "naucs_at_1_diff1": 0.8552027182918394, "naucs_at_3_max": 0.5629191070367531, "naucs_at_3_std": 0.27028690263984145, "naucs_at_3_diff1": 0.7784143960614534, "naucs_at_5_max": 0.5055688942243579, "naucs_at_5_std": 0.46521942110177583, "naucs_at_5_diff1": 0.733126583966918, "naucs_at_10_max": 0.4561741363212023, "naucs_at_10_std": 0.4275793650793735, "naucs_at_10_diff1": 0.5840336134453762, "naucs_at_20_max": 0.24964985994398026, "naucs_at_20_std": 0.24556489262372797, "naucs_at_20_diff1": 0.6377217553688285, "naucs_at_50_max": 0.8692810457515607, "naucs_at_50_std": 1.0, "naucs_at_50_diff1": 0.7222222222221809, "naucs_at_100_max": 1.0, "naucs_at_100_std": 1.0, "naucs_at_100_diff1": 1.0}}