Upload results.json

results.json

@@ -1 +1 @@
-{"validation_set": {"ndcg_at_1": 0.796, "ndcg_at_3": 0.85495, "ndcg_at_5": 0.8639, "ndcg_at_10": 0.87254, "ndcg_at_20": 0.8799, "ndcg_at_50": 0.88348, "ndcg_at_100": 0.88382, "map_at_1": 0.796, "map_at_3": 0.84133, "map_at_5": 0.84623, "map_at_10": 0.84994, "map_at_20": 0.85185, "map_at_50": 0.85242, "map_at_100": 0.85246, "recall_at_1": 0.796, "recall_at_3": 0.894, "recall_at_5": 0.916, "recall_at_10": 0.942, "recall_at_20": 0.972, "recall_at_50": 0.99, "recall_at_100": 0.992, "precision_at_1": 0.796, "precision_at_3": 0.298, "precision_at_5": 0.1832, "precision_at_10": 0.0942, "precision_at_20": 0.0486, "precision_at_50": 0.0198, "precision_at_100": 0.00992, "mrr_at_1": 0.796, "mrr_at_3": 0.843, "mrr_at_5": 0.8469, "mrr_at_10": 0.8505325396825396, "mrr_at_20": 0.8521482558635035, "mrr_at_50": 0.8528272330920051, "mrr_at_100": 0.85286113139709, "naucs_at_1_max": 0.12337304593347541, "naucs_at_1_std": -0.7591714462797723, "naucs_at_1_diff1": 0.9149015711235945, "naucs_at_3_max": -0.012867614433025963, "naucs_at_3_std": -0.9498251779280444, "naucs_at_3_diff1": 0.8684397351933689, "naucs_at_5_max": -0.014372415632921569, "naucs_at_5_std": -0.8904117202436561, "naucs_at_5_diff1": 0.8718932017251367, "naucs_at_10_max": -0.03784732283718035, "naucs_at_10_std": -0.9889565021410871, "naucs_at_10_diff1": 0.8798254934157582, "naucs_at_20_max": 0.08703481392556274, "naucs_at_20_std": -0.838368680805674, "naucs_at_20_diff1": 0.8572762438308682, "naucs_at_50_max": -0.2536881419234251, "naucs_at_50_std": -0.956676003734828, "naucs_at_50_diff1": 0.9183006535947666, "naucs_at_100_max": -0.17623716153129299, "naucs_at_100_std": -0.7608543417366961, "naucs_at_100_diff1": 0.9305555555555584}, "syntheticDocQA_energy": {"ndcg_at_1": 0.87, "ndcg_at_3": 0.91917, "ndcg_at_5": 0.92734, "ndcg_at_10": 0.92734, "ndcg_at_20": 0.93276, "ndcg_at_50": 0.93494, "ndcg_at_100": 0.93494, "map_at_1": 0.87, "map_at_3": 0.90833, "map_at_5": 0.91283, "map_at_10": 0.91283, "map_at_20": 0.91451, "map_at_50": 0.91495, "map_at_100": 0.91495, "recall_at_1": 0.87, "recall_at_3": 0.95, "recall_at_5": 0.97, "recall_at_10": 0.97, "recall_at_20": 0.99, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.87, "precision_at_3": 0.31667, "precision_at_5": 0.194, "precision_at_10": 0.097, "precision_at_20": 0.0495, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.87, "mrr_at_3": 0.9083333333333333, "mrr_at_5": 0.9128333333333333, "mrr_at_10": 0.9139444444444444, "mrr_at_20": 0.9147777777777777, "mrr_at_50": 0.9152323232323232, "mrr_at_100": 0.9152323232323232, "naucs_at_1_max": 0.43582907502036694, "naucs_at_1_std": -0.2828630674664893, "naucs_at_1_diff1": 0.8985410649485305, "naucs_at_3_max": 0.13706816059757365, "naucs_at_3_std": -0.7007469654528471, "naucs_at_3_diff1": 0.9183006535947692, "naucs_at_5_max": -0.3020541549953327, "naucs_at_5_std": -0.7268907563025226, "naucs_at_5_diff1": 0.9074074074074108, "naucs_at_10_max": -0.3020541549953327, "naucs_at_10_std": -0.7268907563025226, "naucs_at_10_diff1": 0.9074074074074108, "naucs_at_20_max": -0.1713352007469681, "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}, "syntheticDocQA_healthcare_industry": {"ndcg_at_1": 0.93, "ndcg_at_3": 0.97286, "ndcg_at_5": 0.97286, "ndcg_at_10": 0.97286, "ndcg_at_20": 0.97286, "ndcg_at_50": 0.97286, "ndcg_at_100": 0.97286, "map_at_1": 0.93, "map_at_3": 0.96333, "map_at_5": 0.96333, "map_at_10": 0.96333, "map_at_20": 0.96333, "map_at_50": 0.96333, "map_at_100": 0.96333, "recall_at_1": 0.93, "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": 0.93, "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": 0.94, "mrr_at_3": 0.9683333333333333, "mrr_at_5": 0.9683333333333333, "mrr_at_10": 0.9683333333333333, "mrr_at_20": 0.9683333333333333, "mrr_at_50": 0.9683333333333333, "mrr_at_100": 0.9683333333333333, "naucs_at_1_max": 0.6639322395624917, "naucs_at_1_std": 0.057823129251700287, "naucs_at_1_diff1": 0.943977591036415, "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": 0.92, "ndcg_at_3": 0.96655, "ndcg_at_5": 0.96655, "ndcg_at_10": 0.96655, "ndcg_at_20": 0.96655, "ndcg_at_50": 0.96655, "ndcg_at_100": 0.96655, "map_at_1": 0.92, "map_at_3": 0.955, "map_at_5": 0.955, "map_at_10": 0.955, "map_at_20": 0.955, "map_at_50": 0.955, "map_at_100": 0.955, "recall_at_1": 0.92, "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": 0.92, "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": 0.92, "mrr_at_3": 0.955, "mrr_at_5": 0.955, "mrr_at_10": 0.955, "mrr_at_20": 0.955, "mrr_at_50": 0.955, "mrr_at_100": 0.955, "naucs_at_1_max": 0.37511671335200736, "naucs_at_1_std": -0.2263655462184869, "naucs_at_1_diff1": 0.9509803921568633, "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": 0.93, "ndcg_at_3": 0.96024, "ndcg_at_5": 0.96454, "ndcg_at_10": 0.96755, "ndcg_at_20": 0.96755, "ndcg_at_50": 0.96755, "ndcg_at_100": 0.96755, "map_at_1": 0.93, "map_at_3": 0.95333, "map_at_5": 0.95583, "map_at_10": 0.95694, "map_at_20": 0.95694, "map_at_50": 0.95694, "map_at_100": 0.95694, "recall_at_1": 0.93, "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": 0.93, "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": 0.93, "mrr_at_3": 0.9533333333333333, "mrr_at_5": 0.9558333333333333, "mrr_at_10": 0.9570833333333333, "mrr_at_20": 0.9570833333333333, "mrr_at_50": 0.9570833333333333, "mrr_at_100": 0.9570833333333333, "naucs_at_1_max": 0.49593170601574016, "naucs_at_1_std": -0.1608643457382955, "naucs_at_1_diff1": 0.9229691876750709, "naucs_at_3_max": 0.4561157796451816, "naucs_at_3_std": -0.661531279178345, "naucs_at_3_diff1": 0.8611111111111119, "naucs_at_5_max": 0.35807656395891135, "naucs_at_5_std": -1.1517273576097316, "naucs_at_5_diff1": 0.7222222222222276, "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": 0.81, "ndcg_at_3": 0.85307, "ndcg_at_5": 0.86873, "ndcg_at_10": 0.87921, "ndcg_at_20": 0.88229, "ndcg_at_50": 0.88317, "ndcg_at_100": 0.88512, "map_at_1": 0.81, "map_at_3": 0.843, "map_at_5": 0.8517, "map_at_10": 0.85609, "map_at_20": 0.85695, "map_at_50": 0.85714, "map_at_100": 0.85731, "recall_at_1": 0.81, "recall_at_3": 0.882, "recall_at_5": 0.92, "recall_at_10": 0.952, "recall_at_20": 0.964, "recall_at_50": 0.968, "recall_at_100": 0.98, "precision_at_1": 0.81, "precision_at_3": 0.294, "precision_at_5": 0.184, "precision_at_10": 0.0952, "precision_at_20": 0.0482, "precision_at_50": 0.01936, "precision_at_100": 0.0098, "mrr_at_1": 0.806, "mrr_at_3": 0.8413333333333334, "mrr_at_5": 0.8497333333333333, "mrr_at_10": 0.8544809523809525, "mrr_at_20": 0.8553121794871795, "mrr_at_50": 0.8554983266733267, "mrr_at_100": 0.8556658414885311, "naucs_at_1_max": 0.141154591649641, "naucs_at_1_std": -0.07635811200167668, "naucs_at_1_diff1": 0.8810571533343818, "naucs_at_3_max": 0.07860614574755015, "naucs_at_3_std": -0.13010867283501834, "naucs_at_3_diff1": 0.8522905262477993, "naucs_at_5_max": 0.2155112044817942, "naucs_at_5_std": 0.259243697478991, "naucs_at_5_diff1": 0.833158263305322, "naucs_at_10_max": 0.17501167133520013, "naucs_at_10_std": 0.5606520385932124, "naucs_at_10_diff1": 0.8140367258014324, "naucs_at_20_max": 0.04276895943562585, "naucs_at_20_std": 0.7092800082996177, "naucs_at_20_diff1": 0.8242815644776446, "naucs_at_50_max": 0.06576797385620768, "naucs_at_50_std": 0.8523576097105556, "naucs_at_50_diff1": 0.8104866946778694, "naucs_at_100_max": 0.17852474323061385, "naucs_at_100_std": 0.8818860877684281, "naucs_at_100_diff1": 0.8921568627450945}, "docvqa_subsampled": {"ndcg_at_1": 0.41, "ndcg_at_3": 0.49335, "ndcg_at_5": 0.51229, "ndcg_at_10": 0.5322, "ndcg_at_20": 0.54738, "ndcg_at_50": 0.56385, "ndcg_at_100": 0.5746, "map_at_1": 0.41, "map_at_3": 0.47367, "map_at_5": 0.48417, "map_at_10": 0.4923, "map_at_20": 0.49649, "map_at_50": 0.49905, "map_at_100": 0.50001, "recall_at_1": 0.41, "recall_at_3": 0.55, "recall_at_5": 0.596, "recall_at_10": 0.658, "recall_at_20": 0.718, "recall_at_50": 0.802, "recall_at_100": 0.868, "precision_at_1": 0.41, "precision_at_3": 0.18333, "precision_at_5": 0.1192, "precision_at_10": 0.0658, "precision_at_20": 0.0359, "precision_at_50": 0.01604, "precision_at_100": 0.00868, "mrr_at_1": 0.41, "mrr_at_3": 0.4726666666666667, "mrr_at_5": 0.48296666666666666, "mrr_at_10": 0.4917436507936508, "mrr_at_20": 0.4959952597969626, "mrr_at_50": 0.4985024962861558, "mrr_at_100": 0.4994159027267983, "naucs_at_1_max": -0.4636449358667992, "naucs_at_1_std": -0.164126742188408, "naucs_at_1_diff1": 0.8276011728690081, "naucs_at_3_max": -0.537999017061012, "naucs_at_3_std": -0.1264452713613703, "naucs_at_3_diff1": 0.741461770694376, "naucs_at_5_max": -0.561941176122966, "naucs_at_5_std": -0.10844064130672434, "naucs_at_5_diff1": 0.7018768799825076, "naucs_at_10_max": -0.5871388568387022, "naucs_at_10_std": -0.07507678269752781, "naucs_at_10_diff1": 0.6741801856008889, "naucs_at_20_max": -0.6113258505388312, "naucs_at_20_std": 0.039377605480469854, "naucs_at_20_diff1": 0.6472060904232113, "naucs_at_50_max": -0.5296614869566092, "naucs_at_50_std": 0.27499016829856143, "naucs_at_50_diff1": 0.6249684881364131, "naucs_at_100_max": -0.645501563276161, "naucs_at_100_std": 0.3408453379306334, "naucs_at_100_diff1": 0.5860021038482894}, "arxivqa_subsampled": {"ndcg_at_1": 0.754, "ndcg_at_3": 0.82374, "ndcg_at_5": 0.83295, "ndcg_at_10": 0.84664, "ndcg_at_20": 0.85383, "ndcg_at_50": 0.85702, "ndcg_at_100": 0.85905, "map_at_1": 0.754, "map_at_3": 0.807, "map_at_5": 0.8122, "map_at_10": 0.81754, "map_at_20": 0.81957, "map_at_50": 0.82009, "map_at_100": 0.82029, "recall_at_1": 0.754, "recall_at_3": 0.872, "recall_at_5": 0.894, "recall_at_10": 0.938, "recall_at_20": 0.966, "recall_at_50": 0.982, "recall_at_100": 0.994, "precision_at_1": 0.754, "precision_at_3": 0.29067, "precision_at_5": 0.1788, "precision_at_10": 0.0938, "precision_at_20": 0.0483, "precision_at_50": 0.01964, "precision_at_100": 0.00994, "mrr_at_1": 0.756, "mrr_at_3": 0.808, "mrr_at_5": 0.8135, "mrr_at_10": 0.8190404761904762, "mrr_at_20": 0.821144196446828, "mrr_at_50": 0.8216980992057245, "mrr_at_100": 0.8218573901238074, "naucs_at_1_max": 0.1428444191856007, "naucs_at_1_std": -0.43192199541099513, "naucs_at_1_diff1": 0.9350541392224136, "naucs_at_3_max": 0.0768953502342892, "naucs_at_3_std": -0.5499294124714647, "naucs_at_3_diff1": 0.8804893067403577, "naucs_at_5_max": 0.05632465434925257, "naucs_at_5_std": -0.48102692437392636, "naucs_at_5_diff1": 0.8673304580870743, "naucs_at_10_max": -0.09901810186440438, "naucs_at_10_std": -0.5959308454564681, "naucs_at_10_diff1": 0.8651396042287877, "naucs_at_20_max": -0.023672213983635542, "naucs_at_20_std": -0.5235349041577391, "naucs_at_20_diff1": 0.894957983193281, "naucs_at_50_max": -0.19664902998236666, "naucs_at_50_std": -0.1808278867102445, "naucs_at_50_diff1": 0.865649963689171, "naucs_at_100_max": -0.9556489262372053, "naucs_at_100_std": 0.14285714285712342, "naucs_at_100_diff1": 0.8692810457516253}, "tabfquad_subsampled": {"ndcg_at_1": 0.88571, "ndcg_at_3": 0.93699, "ndcg_at_5": 0.93699, "ndcg_at_10": 0.94051, "ndcg_at_20": 0.94331, "ndcg_at_50": 0.94405, "ndcg_at_100": 0.94466, "map_at_1": 0.88571, "map_at_3": 0.925, "map_at_5": 0.925, "map_at_10": 0.92649, "map_at_20": 0.9273, "map_at_50": 0.92743, "map_at_100": 0.92749, "recall_at_1": 0.88571, "recall_at_3": 0.97143, "recall_at_5": 0.97143, "recall_at_10": 0.98214, "recall_at_20": 0.99286, "recall_at_50": 0.99643, "recall_at_100": 1.0, "precision_at_1": 0.88571, "precision_at_3": 0.32381, "precision_at_5": 0.19429, "precision_at_10": 0.09821, "precision_at_20": 0.04964, "precision_at_50": 0.01993, "precision_at_100": 0.01, "mrr_at_1": 0.8857142857142857, "mrr_at_3": 0.9255952380952381, "mrr_at_5": 0.9255952380952381, "mrr_at_10": 0.9270833333333333, "mrr_at_20": 0.9279336734693877, "mrr_at_50": 0.9280659486016629, "mrr_at_100": 0.9281264812893142, "naucs_at_1_max": 0.6374288087327964, "naucs_at_1_std": 0.12283459895586181, "naucs_at_1_diff1": 0.8410803274798299, "naucs_at_3_max": 0.8978758169934691, "naucs_at_3_std": 0.2625466853408014, "naucs_at_3_diff1": 0.8631535947712446, "naucs_at_5_max": 0.8978758169934646, "naucs_at_5_std": 0.26254668534080566, "naucs_at_5_diff1": 0.8631535947712403, "naucs_at_10_max": 0.9183006535947714, "naucs_at_10_std": 0.314752567693746, "naucs_at_10_diff1": 0.8071895424836623, "naucs_at_20_max": 0.8611111111111035, "naucs_at_20_std": 0.8611111111111035, "naucs_at_20_diff1": 0.8611111111111035, "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}, "tatdqa": {"ndcg_at_1": 0.52977, "ndcg_at_3": 0.64311, "ndcg_at_5": 0.67365, "ndcg_at_10": 0.70159, "ndcg_at_20": 0.71447, "ndcg_at_50": 0.72263, "ndcg_at_100": 0.72448, "map_at_1": 0.52977, "map_at_3": 0.61535, "map_at_5": 0.63234, "map_at_10": 0.64407, "map_at_20": 0.64766, "map_at_50": 0.64904, "map_at_100": 0.6492, "recall_at_1": 0.52977, "recall_at_3": 0.72339, "recall_at_5": 0.79735, "recall_at_10": 0.88274, "recall_at_20": 0.93325, "recall_at_50": 0.97354, "recall_at_100": 0.98497, "precision_at_1": 0.52977, "precision_at_3": 0.24113, "precision_at_5": 0.15947, "precision_at_10": 0.08827, "precision_at_20": 0.04666, "precision_at_50": 0.01947, "precision_at_100": 0.00985, "mrr_at_1": 0.5279615153337343, "mrr_at_3": 0.6150531168570856, "mrr_at_5": 0.6313189015834837, "mrr_at_10": 0.6434737374605083, "mrr_at_20": 0.6468848430310473, "mrr_at_50": 0.6482633204782727, "mrr_at_100": 0.6484247424510643, "naucs_at_1_max": -0.006443992420677248, "naucs_at_1_std": -0.18218281794726, "naucs_at_1_diff1": 0.7055457414361879, "naucs_at_3_max": -0.050556539227135475, "naucs_at_3_std": -0.17760630819242695, "naucs_at_3_diff1": 0.6162068297024225, "naucs_at_5_max": -0.019222306216811074, "naucs_at_5_std": -0.10747978499063429, "naucs_at_5_diff1": 0.5849707721609257, "naucs_at_10_max": 0.09621599533046171, "naucs_at_10_std": 0.058637070424872303, "naucs_at_10_diff1": 0.5587252535330388, "naucs_at_20_max": 0.11158948785105664, "naucs_at_20_std": 0.2009836323027148, "naucs_at_20_diff1": 0.5618909216838407, "naucs_at_50_max": 0.1088579108612422, "naucs_at_50_std": 0.7402223975810361, "naucs_at_50_diff1": 0.6047747365371076, "naucs_at_100_max": -0.049397070807442954, "naucs_at_100_std": 0.8340111571777812, "naucs_at_100_diff1": 0.6156531134683879}, "shift_project": {"ndcg_at_1": 0.67, "ndcg_at_3": 0.77464, "ndcg_at_5": 0.82325, "ndcg_at_10": 0.82997, "ndcg_at_20": 0.82997, "ndcg_at_50": 0.83393, "ndcg_at_100": 0.83393, "map_at_1": 0.67, "map_at_3": 0.75167, "map_at_5": 0.77817, "map_at_10": 0.78108, "map_at_20": 0.78108, "map_at_50": 0.78171, "map_at_100": 0.78171, "recall_at_1": 0.67, "recall_at_3": 0.84, "recall_at_5": 0.96, "recall_at_10": 0.98, "recall_at_20": 0.98, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.67, "precision_at_3": 0.28, "precision_at_5": 0.192, "precision_at_10": 0.098, "precision_at_20": 0.049, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.66, "mrr_at_3": 0.75, "mrr_at_5": 0.773, "mrr_at_10": 0.7775833333333334, "mrr_at_20": 0.7775833333333334, "mrr_at_50": 0.778210113960114, "mrr_at_100": 0.778210113960114, "naucs_at_1_max": 0.32210390885544504, "naucs_at_1_std": -0.3513568417666881, "naucs_at_1_diff1": 0.6506068967852247, "naucs_at_3_max": 0.4938328424153158, "naucs_at_3_std": -0.3930412371134031, "naucs_at_3_diff1": 0.6315660284732447, "naucs_at_5_max": 0.658613445378152, "naucs_at_5_std": 0.2685574229691903, "naucs_at_5_diff1": 0.658613445378152, "naucs_at_10_max": 0.861111111111116, "naucs_at_10_std": -0.024276377217554025, "naucs_at_10_diff1": 0.540149393090577, "naucs_at_20_max": 0.861111111111116, "naucs_at_20_std": -0.024276377217554025, "naucs_at_20_diff1": 0.540149393090577, "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": 0.53524, "ndcg_at_3": 0.648, "ndcg_at_5": 0.67885, "ndcg_at_10": 0.70688, "ndcg_at_20": 0.71942, "ndcg_at_50": 0.72753, "ndcg_at_100": 0.7292, "map_at_1": 0.53524, "map_at_3": 0.62039, "map_at_5": 0.63756, "map_at_10": 0.64932, "map_at_20": 0.65281, "map_at_50": 0.65418, "map_at_100": 0.65432, "recall_at_1": 0.53524, "recall_at_3": 0.72783, "recall_at_5": 0.80255, "recall_at_10": 0.88821, "recall_at_20": 0.93742, "recall_at_50": 0.97752, "recall_at_100": 0.98785, "precision_at_1": 0.53524, "precision_at_3": 0.24261, "precision_at_5": 0.16051, "precision_at_10": 0.08882, "precision_at_20": 0.04687, "precision_at_50": 0.01955, "precision_at_100": 0.00988, "mrr_at_1": 0.5334143377885784, "mrr_at_3": 0.6202916160388822, "mrr_at_5": 0.6365735115431348, "mrr_at_10": 0.6487063395629616, "mrr_at_20": 0.6520654396967506, "mrr_at_50": 0.6534292239902603, "mrr_at_100": 0.6535764192591313, "naucs_at_1_max": 0.0070613685277818505, "naucs_at_1_std": -0.19360409606847048, "naucs_at_1_diff1": 0.7055038908717195, "naucs_at_3_max": -0.03338657220104696, "naucs_at_3_std": -0.1963193270085075, "naucs_at_3_diff1": 0.6152079200615186, "naucs_at_5_max": 0.00978659323130093, "naucs_at_5_std": -0.13077523724101206, "naucs_at_5_diff1": 0.5803061735651573, "naucs_at_10_max": 0.14712007460201967, "naucs_at_10_std": 0.013885140384765186, "naucs_at_10_diff1": 0.5362333351419983, "naucs_at_20_max": 0.19393798047158722, "naucs_at_20_std": 0.14038843262382952, "naucs_at_20_diff1": 0.5335539615922209, "naucs_at_50_max": 0.3348474540218742, "naucs_at_50_std": 0.6945619412923183, "naucs_at_50_diff1": 0.5322399798882755, "naucs_at_100_max": 0.19655877606932606, "naucs_at_100_std": 0.7925240243357159, "naucs_at_100_diff1": 0.5097832357595132}, "../colpali/data_dir/eval_vidore/shiftproject_test": {"ndcg_at_1": 0.67, "ndcg_at_3": 0.77464, "ndcg_at_5": 0.82325, "ndcg_at_10": 0.82997, "ndcg_at_20": 0.82997, "ndcg_at_50": 0.83393, "ndcg_at_100": 0.83393, "map_at_1": 0.67, "map_at_3": 0.75167, "map_at_5": 0.77817, "map_at_10": 0.78108, "map_at_20": 0.78108, "map_at_50": 0.78171, "map_at_100": 0.78171, "recall_at_1": 0.67, "recall_at_3": 0.84, "recall_at_5": 0.96, "recall_at_10": 0.98, "recall_at_20": 0.98, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.67, "precision_at_3": 0.28, "precision_at_5": 0.192, "precision_at_10": 0.098, "precision_at_20": 0.049, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.66, "mrr_at_3": 0.75, "mrr_at_5": 0.773, "mrr_at_10": 0.7775833333333334, "mrr_at_20": 0.7775833333333334, "mrr_at_50": 0.778210113960114, "mrr_at_100": 0.778210113960114, "naucs_at_1_max": 0.32210390885544504, "naucs_at_1_std": -0.3513568417666881, "naucs_at_1_diff1": 0.6506068967852247, "naucs_at_3_max": 0.4938328424153158, "naucs_at_3_std": -0.3930412371134031, "naucs_at_3_diff1": 0.6315660284732447, "naucs_at_5_max": 0.658613445378152, "naucs_at_5_std": 0.2685574229691903, "naucs_at_5_diff1": 0.658613445378152, "naucs_at_10_max": 0.861111111111116, "naucs_at_10_std": -0.024276377217554025, "naucs_at_10_diff1": 0.540149393090577, "naucs_at_20_max": 0.861111111111116, "naucs_at_20_std": -0.024276377217554025, "naucs_at_20_diff1": 0.540149393090577, "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": 0.92, "ndcg_at_3": 0.96655, "ndcg_at_5": 0.96655, "ndcg_at_10": 0.96655, "ndcg_at_20": 0.96655, "ndcg_at_50": 0.96655, "ndcg_at_100": 0.96655, "map_at_1": 0.92, "map_at_3": 0.955, "map_at_5": 0.955, "map_at_10": 0.955, "map_at_20": 0.955, "map_at_50": 0.955, "map_at_100": 0.955, "recall_at_1": 0.92, "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": 0.92, "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": 0.92, "mrr_at_3": 0.955, "mrr_at_5": 0.955, "mrr_at_10": 0.955, "mrr_at_20": 0.955, "mrr_at_50": 0.955, "mrr_at_100": 0.955, "naucs_at_1_max": 0.37511671335200736, "naucs_at_1_std": -0.2263655462184869, "naucs_at_1_diff1": 0.9509803921568633, "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": 0.93, "ndcg_at_3": 0.96024, "ndcg_at_5": 0.96454, "ndcg_at_10": 0.96755, "ndcg_at_20": 0.96755, "ndcg_at_50": 0.96755, "ndcg_at_100": 0.96755, "map_at_1": 0.93, "map_at_3": 0.95333, "map_at_5": 0.95583, "map_at_10": 0.95694, "map_at_20": 0.95694, "map_at_50": 0.95694, "map_at_100": 0.95694, "recall_at_1": 0.93, "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": 0.93, "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": 0.93, "mrr_at_3": 0.9533333333333333, "mrr_at_5": 0.9558333333333333, "mrr_at_10": 0.9570833333333333, "mrr_at_20": 0.9570833333333333, "mrr_at_50": 0.9570833333333333, "mrr_at_100": 0.9570833333333333, "naucs_at_1_max": 0.49593170601574016, "naucs_at_1_std": -0.1608643457382955, "naucs_at_1_diff1": 0.9229691876750709, "naucs_at_3_max": 0.4561157796451816, "naucs_at_3_std": -0.661531279178345, "naucs_at_3_diff1": 0.8611111111111119, "naucs_at_5_max": 0.35807656395891135, "naucs_at_5_std": -1.1517273576097316, "naucs_at_5_diff1": 0.7222222222222276, "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": 0.754, "ndcg_at_3": 0.82374, "ndcg_at_5": 0.83295, "ndcg_at_10": 0.84664, "ndcg_at_20": 0.85383, "ndcg_at_50": 0.85703, "ndcg_at_100": 0.85905, "map_at_1": 0.754, "map_at_3": 0.807, "map_at_5": 0.8122, "map_at_10": 0.81754, "map_at_20": 0.81957, "map_at_50": 0.82009, "map_at_100": 0.8203, "recall_at_1": 0.754, "recall_at_3": 0.872, "recall_at_5": 0.894, "recall_at_10": 0.938, "recall_at_20": 0.966, "recall_at_50": 0.982, "recall_at_100": 0.994, "precision_at_1": 0.754, "precision_at_3": 0.29067, "precision_at_5": 0.1788, "precision_at_10": 0.0938, "precision_at_20": 0.0483, "precision_at_50": 0.01964, "precision_at_100": 0.00994, "mrr_at_1": 0.756, "mrr_at_3": 0.808, "mrr_at_5": 0.8135, "mrr_at_10": 0.8188404761904763, "mrr_at_20": 0.8211260146286462, "mrr_at_50": 0.8216799173875426, "mrr_at_100": 0.8218395169476009, "naucs_at_1_max": 0.1500714646218314, "naucs_at_1_std": -0.43192199541099513, "naucs_at_1_diff1": 0.9337181537104351, "naucs_at_3_max": 0.08698786495253977, "naucs_at_3_std": -0.5499294124714654, "naucs_at_3_diff1": 0.8828547398774474, "naucs_at_5_max": 0.06825160179614095, "naucs_at_5_std": -0.48102692437392636, "naucs_at_5_diff1": 0.8701258363949383, "naucs_at_10_max": -0.07877774765820478, "naucs_at_10_std": -0.5959308454564681, "naucs_at_10_diff1": 0.8746272702629423, "naucs_at_20_max": -0.009831383533804266, "naucs_at_20_std": -0.5235349041577391, "naucs_at_20_diff1": 0.894957983193281, "naucs_at_50_max": -0.17050523913268975, "naucs_at_50_std": -0.1808278867102445, "naucs_at_50_diff1": 0.8801742919389941, "naucs_at_100_max": -0.9556489262372053, "naucs_at_100_std": 0.14285714285712342, "naucs_at_100_diff1": 0.9128540305010608}, "../colpali/data_dir/eval_vidore/docvqa_test_subsampled": {"ndcg_at_1": 0.44789, "ndcg_at_3": 0.5378, "ndcg_at_5": 0.55783, "ndcg_at_10": 0.57699, "ndcg_at_20": 0.59148, "ndcg_at_50": 0.60746, "ndcg_at_100": 0.6165, "map_at_1": 0.44789, "map_at_3": 0.51663, "map_at_5": 0.52772, "map_at_10": 0.5355, "map_at_20": 0.53944, "map_at_50": 0.54189, "map_at_100": 0.54269, "recall_at_1": 0.44789, "recall_at_3": 0.59867, "recall_at_5": 0.64745, "recall_at_10": 0.70732, "recall_at_20": 0.76497, "recall_at_50": 0.84701, "recall_at_100": 0.90244, "precision_at_1": 0.44789, "precision_at_3": 0.19956, "precision_at_5": 0.12949, "precision_at_10": 0.07073, "precision_at_20": 0.03825, "precision_at_50": 0.01694, "precision_at_100": 0.00902, "mrr_at_1": 0.44789356984478934, "mrr_at_3": 0.5155210643015521, "mrr_at_5": 0.5263858093126386, "mrr_at_10": 0.5349311934677788, "mrr_at_20": 0.5389739011588118, "mrr_at_50": 0.5413416879514581, "mrr_at_100": 0.542101189253761, "naucs_at_1_max": -0.4565514230867296, "naucs_at_1_std": -0.18254253065347953, "naucs_at_1_diff1": 0.8416832179210567, "naucs_at_3_max": -0.5449065298373567, "naucs_at_3_std": -0.15643575661002856, "naucs_at_3_diff1": 0.7336307024005878, "naucs_at_5_max": -0.5812342843487461, "naucs_at_5_std": -0.14516981817744357, "naucs_at_5_diff1": 0.6939261968512778, "naucs_at_10_max": -0.6344654602451121, "naucs_at_10_std": -0.12596239599831063, "naucs_at_10_diff1": 0.6471521211864805, "naucs_at_20_max": -0.6938965921633944, "naucs_at_20_std": -0.01757178727905947, "naucs_at_20_diff1": 0.620742671327756, "naucs_at_50_max": -0.6116871360480838, "naucs_at_50_std": 0.26802002565467087, "naucs_at_50_diff1": 0.6009703615343458, "naucs_at_100_max": -0.7203826564997696, "naucs_at_100_std": 0.3529932440739904, "naucs_at_100_diff1": 0.5353099546350089}, "../colpali/data_dir/eval_vidore/syntheticDocQA_healthcare_industry_test": {"ndcg_at_1": 0.93, "ndcg_at_3": 0.97286, "ndcg_at_5": 0.97286, "ndcg_at_10": 0.97286, "ndcg_at_20": 0.97286, "ndcg_at_50": 0.97286, "ndcg_at_100": 0.97286, "map_at_1": 0.93, "map_at_3": 0.96333, "map_at_5": 0.96333, "map_at_10": 0.96333, "map_at_20": 0.96333, "map_at_50": 0.96333, "map_at_100": 0.96333, "recall_at_1": 0.93, "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": 0.93, "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": 0.94, "mrr_at_3": 0.9683333333333333, "mrr_at_5": 0.9683333333333333, "mrr_at_10": 0.9683333333333333, "mrr_at_20": 0.9683333333333333, "mrr_at_50": 0.9683333333333333, "mrr_at_100": 0.9683333333333333, "naucs_at_1_max": 0.6639322395624917, "naucs_at_1_std": 0.057823129251700287, "naucs_at_1_diff1": 0.943977591036415, "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/infovqa_test_subsampled": {"ndcg_at_1": 0.81984, "ndcg_at_3": 0.86343, "ndcg_at_5": 0.87929, "ndcg_at_10": 0.88922, "ndcg_at_20": 0.89233, "ndcg_at_50": 0.89322, "ndcg_at_100": 0.89519, "map_at_1": 0.81984, "map_at_3": 0.85324, "map_at_5": 0.86204, "map_at_10": 0.8662, "map_at_20": 0.86707, "map_at_50": 0.86726, "map_at_100": 0.86743, "recall_at_1": 0.81984, "recall_at_3": 0.89271, "recall_at_5": 0.93117, "recall_at_10": 0.96154, "recall_at_20": 0.97368, "recall_at_50": 0.97773, "recall_at_100": 0.98988, "precision_at_1": 0.81984, "precision_at_3": 0.29757, "precision_at_5": 0.18623, "precision_at_10": 0.09615, "precision_at_20": 0.04868, "precision_at_50": 0.01955, "precision_at_100": 0.0099, "mrr_at_1": 0.8157894736842105, "mrr_at_3": 0.8515519568151148, "mrr_at_5": 0.8600539811066127, "mrr_at_10": 0.8645700790437633, "mrr_at_20": 0.8654114020257745, "mrr_at_50": 0.8655998101089194, "mrr_at_100": 0.865769359517021, "naucs_at_1_max": 0.14660094857145425, "naucs_at_1_std": -0.10725255805767692, "naucs_at_1_diff1": 0.8848448569449995, "naucs_at_3_max": 0.07815385670883496, "naucs_at_3_std": -0.18709885833066617, "naucs_at_3_diff1": 0.8516996021765253, "naucs_at_5_max": 0.24070807197229796, "naucs_at_5_std": 0.23830762281738507, "naucs_at_5_diff1": 0.8452353335313445, "naucs_at_10_max": 0.12987525811013112, "naucs_at_10_std": 0.4853319241865675, "naucs_at_10_diff1": 0.8238624110525148, "naucs_at_20_max": -0.07404577937479512, "naucs_at_20_std": 0.6566492096062518, "naucs_at_20_diff1": 0.8296278119647126, "naucs_at_50_max": -0.06194492068530538, "naucs_at_50_std": 0.8555161767364838, "naucs_at_50_diff1": 0.8105239099641848, "naucs_at_100_max": 0.009627852205461772, "naucs_at_100_std": 0.9183327616354471, "naucs_at_100_diff1": 0.9444530524483902}, "../colpali/data_dir/eval_vidore/syntheticDocQA_energy_test": {"ndcg_at_1": 0.87, "ndcg_at_3": 0.91917, "ndcg_at_5": 0.92734, "ndcg_at_10": 0.92734, "ndcg_at_20": 0.93276, "ndcg_at_50": 0.93494, "ndcg_at_100": 0.93494, "map_at_1": 0.87, "map_at_3": 0.90833, "map_at_5": 0.91283, "map_at_10": 0.91283, "map_at_20": 0.91451, "map_at_50": 0.91495, "map_at_100": 0.91495, "recall_at_1": 0.87, "recall_at_3": 0.95, "recall_at_5": 0.97, "recall_at_10": 0.97, "recall_at_20": 0.99, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.87, "precision_at_3": 0.31667, "precision_at_5": 0.194, "precision_at_10": 0.097, "precision_at_20": 0.0495, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.87, "mrr_at_3": 0.9083333333333333, "mrr_at_5": 0.9128333333333333, "mrr_at_10": 0.9139444444444444, "mrr_at_20": 0.9147777777777777, "mrr_at_50": 0.9152323232323232, "mrr_at_100": 0.9152323232323232, "naucs_at_1_max": 0.43582907502036694, "naucs_at_1_std": -0.2828630674664893, "naucs_at_1_diff1": 0.8985410649485305, "naucs_at_3_max": 0.13706816059757365, "naucs_at_3_std": -0.7007469654528471, "naucs_at_3_diff1": 0.9183006535947692, "naucs_at_5_max": -0.3020541549953327, "naucs_at_5_std": -0.7268907563025226, "naucs_at_5_diff1": 0.9074074074074108, "naucs_at_10_max": -0.3020541549953327, "naucs_at_10_std": -0.7268907563025226, "naucs_at_10_diff1": 0.9074074074074108, "naucs_at_20_max": -0.1713352007469681, "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/tabfquad_test_subsampled": {"ndcg_at_1": 0.88571, "ndcg_at_3": 0.93699, "ndcg_at_5": 0.93699, "ndcg_at_10": 0.94051, "ndcg_at_20": 0.94331, "ndcg_at_50": 0.94405, "ndcg_at_100": 0.94466, "map_at_1": 0.88571, "map_at_3": 0.925, "map_at_5": 0.925, "map_at_10": 0.92649, "map_at_20": 0.9273, "map_at_50": 0.92743, "map_at_100": 0.92749, "recall_at_1": 0.88571, "recall_at_3": 0.97143, "recall_at_5": 0.97143, "recall_at_10": 0.98214, "recall_at_20": 0.99286, "recall_at_50": 0.99643, "recall_at_100": 1.0, "precision_at_1": 0.88571, "precision_at_3": 0.32381, "precision_at_5": 0.19429, "precision_at_10": 0.09821, "precision_at_20": 0.04964, "precision_at_50": 0.01993, "precision_at_100": 0.01, "mrr_at_1": 0.8857142857142857, "mrr_at_3": 0.9255952380952381, "mrr_at_5": 0.9255952380952381, "mrr_at_10": 0.9270833333333333, "mrr_at_20": 0.9279336734693877, "mrr_at_50": 0.9280659486016629, "mrr_at_100": 0.9281264812893142, "naucs_at_1_max": 0.6374288087327964, "naucs_at_1_std": 0.12283459895586181, "naucs_at_1_diff1": 0.8410803274798299, "naucs_at_3_max": 0.8978758169934691, "naucs_at_3_std": 0.2625466853408014, "naucs_at_3_diff1": 0.8631535947712446, "naucs_at_5_max": 0.8978758169934646, "naucs_at_5_std": 0.26254668534080566, "naucs_at_5_diff1": 0.8631535947712403, "naucs_at_10_max": 0.9183006535947714, "naucs_at_10_std": 0.314752567693746, "naucs_at_10_diff1": 0.8071895424836623, "naucs_at_20_max": 0.8611111111111035, "naucs_at_20_std": 0.8611111111111035, "naucs_at_20_diff1": 0.8611111111111035, "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}}