Upload folder using huggingface_hub

results.json

@@ -1 +1 @@
-{"validation_set": {"ndcg_at_1": 0.79, "ndcg_at_3": 0.84843, "ndcg_at_5": 0.85669, "ndcg_at_10": 0.86202, "ndcg_at_20": 0.8677, "ndcg_at_50": 0.87297, "ndcg_at_100": 0.87529, "map_at_1": 0.79, "map_at_3": 0.83467, "map_at_5": 0.83927, "map_at_10": 0.84156, "map_at_20": 0.84317, "map_at_50": 0.84407, "map_at_100": 0.84429, "recall_at_1": 0.79, "recall_at_3": 0.888, "recall_at_5": 0.908, "recall_at_10": 0.924, "recall_at_20": 0.946, "recall_at_50": 0.972, "recall_at_100": 0.986, "precision_at_1": 0.79, "precision_at_3": 0.296, "precision_at_5": 0.1816, "precision_at_10": 0.0924, "precision_at_20": 0.0473, "precision_at_50": 0.01944, "precision_at_100": 0.00986, "mrr_at_1": 0.794, "mrr_at_3": 0.8353333333333334, "mrr_at_5": 0.8422333333333334, "mrr_at_10": 0.8441301587301587, "mrr_at_20": 0.8458793785419947, "mrr_at_50": 0.8465229562956768, "mrr_at_100": 0.8467717729113465, "naucs_at_1_max": 0.33967826484237773, "naucs_at_1_std": -0.582754226080265, "naucs_at_1_diff1": 0.9353692259600352, "naucs_at_3_max": 0.4177489177489186, "naucs_at_3_std": -0.6576873647186127, "naucs_at_3_diff1": 0.9209111201298699, "naucs_at_5_max": 0.3537632444282065, "naucs_at_5_std": -0.8584622254698979, "naucs_at_5_diff1": 0.9243494499248974, "naucs_at_10_max": 0.32978279030910307, "naucs_at_10_std": -0.8843063541206009, "naucs_at_10_diff1": 0.9574303405572758, "naucs_at_20_max": 0.24876370301207096, "naucs_at_20_std": -1.1399522772071786, "naucs_at_20_diff1": 0.9806342290002422, "naucs_at_50_max": 0.337434973989592, "naucs_at_50_std": -1.3481726023742884, "naucs_at_50_diff1": 0.9813258636788064, "naucs_at_100_max": 0.3099239695878289, "naucs_at_100_std": -1.2645058023209355, "naucs_at_100_diff1": 1.0}, "syntheticDocQA_energy": {"ndcg_at_1": 0.87, "ndcg_at_3": 0.93047, "ndcg_at_5": 0.93478, "ndcg_at_10": 0.93834, "ndcg_at_20": 0.93834, "ndcg_at_50": 0.94059, "ndcg_at_100": 0.94059, "map_at_1": 0.87, "map_at_3": 0.91667, "map_at_5": 0.91917, "map_at_10": 0.92083, "map_at_20": 0.92083, "map_at_50": 0.92131, "map_at_100": 0.92131, "recall_at_1": 0.87, "recall_at_3": 0.97, "recall_at_5": 0.98, "recall_at_10": 0.99, "recall_at_20": 0.99, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.87, "precision_at_3": 0.32333, "precision_at_5": 0.196, "precision_at_10": 0.099, "precision_at_20": 0.0495, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.89, "mrr_at_3": 0.93, "mrr_at_5": 0.932, "mrr_at_10": 0.932, "mrr_at_20": 0.9325, "mrr_at_50": 0.9325, "mrr_at_100": 0.9325, "naucs_at_1_max": 0.1931052358735098, "naucs_at_1_std": -0.27316151966229746, "naucs_at_1_diff1": 0.975005554321262, "naucs_at_3_max": 0.7424525365701778, "naucs_at_3_std": -0.45238095238094883, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 0.9346405228758136, "naucs_at_5_std": 0.1914098972922579, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 1.0, "naucs_at_10_std": 0.5541549953314738, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 1.0, "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}, "syntheticDocQA_healthcare_industry": {"ndcg_at_1": 0.92, "ndcg_at_3": 0.95655, "ndcg_at_5": 0.96516, "ndcg_at_10": 0.96516, "ndcg_at_20": 0.96516, "ndcg_at_50": 0.96516, "ndcg_at_100": 0.96516, "map_at_1": 0.92, "map_at_3": 0.94833, "map_at_5": 0.95333, "map_at_10": 0.95333, "map_at_20": 0.95333, "map_at_50": 0.95333, "map_at_100": 0.95333, "recall_at_1": 0.92, "recall_at_3": 0.98, "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": 0.32667, "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": 0.9483333333333333, "mrr_at_5": 0.9533333333333333, "mrr_at_10": 0.9533333333333333, "mrr_at_20": 0.9533333333333333, "mrr_at_50": 0.9533333333333333, "mrr_at_100": 0.9533333333333333, "naucs_at_1_max": 0.5192577030812326, "naucs_at_1_std": -0.6149626517273588, "naucs_at_1_diff1": 0.8768674136321195, "naucs_at_3_max": -0.07586367880486825, "naucs_at_3_std": -0.6909430438842241, "naucs_at_3_diff1": 0.7770774976657261, "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": 0.9, "ndcg_at_3": 0.95917, "ndcg_at_5": 0.95917, "ndcg_at_10": 0.95917, "ndcg_at_20": 0.95917, "ndcg_at_50": 0.95917, "ndcg_at_100": 0.95917, "map_at_1": 0.9, "map_at_3": 0.945, "map_at_5": 0.945, "map_at_10": 0.945, "map_at_20": 0.945, "map_at_50": 0.945, "map_at_100": 0.945, "recall_at_1": 0.9, "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": 0.9, "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": 0.9, "mrr_at_3": 0.9483333333333333, "mrr_at_5": 0.9483333333333333, "mrr_at_10": 0.9483333333333333, "mrr_at_20": 0.9483333333333333, "mrr_at_50": 0.9483333333333333, "mrr_at_100": 0.9483333333333333, "naucs_at_1_max": 0.3045751633986925, "naucs_at_1_std": -0.11125116713351993, "naucs_at_1_diff1": 0.9444444444444449, "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": 0.81, "ndcg_at_3": 0.90571, "ndcg_at_5": 0.91389, "ndcg_at_10": 0.91389, "ndcg_at_20": 0.91668, "ndcg_at_50": 0.91668, "ndcg_at_100": 0.91668, "map_at_1": 0.81, "map_at_3": 0.88333, "map_at_5": 0.88783, "map_at_10": 0.88783, "map_at_20": 0.88874, "map_at_50": 0.88874, "map_at_100": 0.88874, "recall_at_1": 0.81, "recall_at_3": 0.97, "recall_at_5": 0.99, "recall_at_10": 0.99, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.81, "precision_at_3": 0.32333, "precision_at_5": 0.198, "precision_at_10": 0.099, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.83, "mrr_at_3": 0.8983333333333333, "mrr_at_5": 0.9003333333333333, "mrr_at_10": 0.9013333333333334, "mrr_at_20": 0.9013333333333334, "mrr_at_50": 0.9013333333333334, "mrr_at_100": 0.9013333333333334, "naucs_at_1_max": 0.44109172822044124, "naucs_at_1_std": -0.10191733459060262, "naucs_at_1_diff1": 0.9325527790874335, "naucs_at_3_max": 0.5720510426392755, "naucs_at_3_std": -0.2759103641456547, "naucs_at_3_diff1": 0.8638344226579548, "naucs_at_5_max": 1.0, "naucs_at_5_std": 0.5541549953314738, "naucs_at_5_diff1": 0.7222222222222276, "naucs_at_10_max": 1.0, "naucs_at_10_std": 0.5541549953314738, "naucs_at_10_diff1": 0.7222222222222276, "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": 0.788, "ndcg_at_3": 0.83459, "ndcg_at_5": 0.8444, "ndcg_at_10": 0.85477, "ndcg_at_20": 0.86156, "ndcg_at_50": 0.86512, "ndcg_at_100": 0.86701, "map_at_1": 0.788, "map_at_3": 0.82367, "map_at_5": 0.82907, "map_at_10": 0.83335, "map_at_20": 0.83533, "map_at_50": 0.83589, "map_at_100": 0.83604, "recall_at_1": 0.788, "recall_at_3": 0.866, "recall_at_5": 0.89, "recall_at_10": 0.922, "recall_at_20": 0.948, "recall_at_50": 0.966, "recall_at_100": 0.978, "precision_at_1": 0.788, "precision_at_3": 0.28867, "precision_at_5": 0.178, "precision_at_10": 0.0922, "precision_at_20": 0.0474, "precision_at_50": 0.01932, "precision_at_100": 0.00978, "mrr_at_1": 0.79, "mrr_at_3": 0.8243333333333334, "mrr_at_5": 0.8296333333333333, "mrr_at_10": 0.8337198412698412, "mrr_at_20": 0.835826794404426, "mrr_at_50": 0.8363718978560812, "mrr_at_100": 0.8364841255520045, "naucs_at_1_max": 0.564645172377579, "naucs_at_1_std": 0.07679418756742774, "naucs_at_1_diff1": 0.9027840100820121, "naucs_at_3_max": 0.6547775199273532, "naucs_at_3_std": 0.20628594490969693, "naucs_at_3_diff1": 0.8634273606526658, "naucs_at_5_max": 0.6722876041577192, "naucs_at_5_std": 0.249360020616785, "naucs_at_5_diff1": 0.8348939094579507, "naucs_at_10_max": 0.8402044578515178, "naucs_at_10_std": 0.488951136009961, "naucs_at_10_diff1": 0.8440230793171982, "naucs_at_20_max": 0.8576276664511968, "naucs_at_20_std": 0.5467751203045319, "naucs_at_20_diff1": 0.8361524096818206, "naucs_at_50_max": 0.9004503762289214, "naucs_at_50_std": 0.7550667325753828, "naucs_at_50_diff1": 0.7987037952435864, "naucs_at_100_max": 0.8866819455054707, "naucs_at_100_std": 0.6872506578388877, "naucs_at_100_diff1": 0.8189882013411434}, "docvqa_subsampled": {"ndcg_at_1": 0.442, "ndcg_at_3": 0.50843, "ndcg_at_5": 0.52685, "ndcg_at_10": 0.54636, "ndcg_at_20": 0.55977, "ndcg_at_50": 0.57411, "ndcg_at_100": 0.58388, "map_at_1": 0.442, "map_at_3": 0.492, "map_at_5": 0.5024, "map_at_10": 0.51014, "map_at_20": 0.51396, "map_at_50": 0.51629, "map_at_100": 0.51716, "recall_at_1": 0.442, "recall_at_3": 0.556, "recall_at_5": 0.6, "recall_at_10": 0.662, "recall_at_20": 0.714, "recall_at_50": 0.786, "recall_at_100": 0.846, "precision_at_1": 0.442, "precision_at_3": 0.18533, "precision_at_5": 0.12, "precision_at_10": 0.0662, "precision_at_20": 0.0357, "precision_at_50": 0.01572, "precision_at_100": 0.00846, "mrr_at_1": 0.44, "mrr_at_3": 0.492, "mrr_at_5": 0.5027, "mrr_at_10": 0.5095595238095239, "mrr_at_20": 0.5132770588458669, "mrr_at_50": 0.5156818249348138, "mrr_at_100": 0.5163464533728304, "naucs_at_1_max": 0.3535234553133254, "naucs_at_1_std": 0.6565424405226743, "naucs_at_1_diff1": 0.8737923667391878, "naucs_at_3_max": 0.25263303691056965, "naucs_at_3_std": 0.7955022132774029, "naucs_at_3_diff1": 0.7975232775726874, "naucs_at_5_max": 0.2028851291184328, "naucs_at_5_std": 0.8435470466013651, "naucs_at_5_diff1": 0.7929504303947756, "naucs_at_10_max": 0.1453567608454435, "naucs_at_10_std": 0.881687287364671, "naucs_at_10_diff1": 0.7690826149688607, "naucs_at_20_max": 0.04475463124388333, "naucs_at_20_std": 0.8884602262289066, "naucs_at_20_diff1": 0.7420375598471808, "naucs_at_50_max": -0.04904539186557294, "naucs_at_50_std": 0.8852432604049678, "naucs_at_50_diff1": 0.7128641087243929, "naucs_at_100_max": -0.023383786394195364, "naucs_at_100_std": 0.870448739944336, "naucs_at_100_diff1": 0.6722266702250679}, "arxivqa_subsampled": {"ndcg_at_1": 0.764, "ndcg_at_3": 0.81433, "ndcg_at_5": 0.82896, "ndcg_at_10": 0.84226, "ndcg_at_20": 0.85147, "ndcg_at_50": 0.85508, "ndcg_at_100": 0.85638, "map_at_1": 0.764, "map_at_3": 0.802, "map_at_5": 0.81, "map_at_10": 0.8157, "map_at_20": 0.8183, "map_at_50": 0.8189, "map_at_100": 0.81901, "recall_at_1": 0.764, "recall_at_3": 0.85, "recall_at_5": 0.886, "recall_at_10": 0.926, "recall_at_20": 0.962, "recall_at_50": 0.98, "recall_at_100": 0.988, "precision_at_1": 0.764, "precision_at_3": 0.28333, "precision_at_5": 0.1772, "precision_at_10": 0.0926, "precision_at_20": 0.0481, "precision_at_50": 0.0196, "precision_at_100": 0.00988, "mrr_at_1": 0.77, "mrr_at_3": 0.8053333333333333, "mrr_at_5": 0.8113333333333334, "mrr_at_10": 0.8185563492063492, "mrr_at_20": 0.820747419771491, "mrr_at_50": 0.8213430501215044, "mrr_at_100": 0.821454277542236, "naucs_at_1_max": 0.671075133964199, "naucs_at_1_std": 0.11374898046885278, "naucs_at_1_diff1": 0.9173193127702304, "naucs_at_3_max": 0.6937845753335508, "naucs_at_3_std": 0.08132769280833158, "naucs_at_3_diff1": 0.8599088838268806, "naucs_at_5_max": 0.7187390746999195, "naucs_at_5_std": 0.10096225881099288, "naucs_at_5_diff1": 0.8398038856610107, "naucs_at_10_max": 0.6757892346127655, "naucs_at_10_std": 0.07752290105231209, "naucs_at_10_diff1": 0.834961011431599, "naucs_at_20_max": 0.866823922551477, "naucs_at_20_std": 0.4288417121234476, "naucs_at_20_diff1": 0.9049093321539146, "naucs_at_50_max": 0.9161998132586351, "naucs_at_50_std": 0.5491129785247385, "naucs_at_50_diff1": 0.8846872082166126, "naucs_at_100_max": 0.9346405228758138, "naucs_at_100_std": 0.7751322751322711, "naucs_at_100_diff1": 0.9256924992219123}, "tabfquad_subsampled": {"ndcg_at_1": 0.77857, "ndcg_at_3": 0.85229, "ndcg_at_5": 0.86705, "ndcg_at_10": 0.87412, "ndcg_at_20": 0.88117, "ndcg_at_50": 0.88249, "ndcg_at_100": 0.8831, "map_at_1": 0.77857, "map_at_3": 0.83452, "map_at_5": 0.84274, "map_at_10": 0.84574, "map_at_20": 0.84758, "map_at_50": 0.84775, "map_at_100": 0.84782, "recall_at_1": 0.77857, "recall_at_3": 0.90357, "recall_at_5": 0.93929, "recall_at_10": 0.96071, "recall_at_20": 0.98929, "recall_at_50": 0.99643, "recall_at_100": 1.0, "precision_at_1": 0.77857, "precision_at_3": 0.30119, "precision_at_5": 0.18786, "precision_at_10": 0.09607, "precision_at_20": 0.04946, "precision_at_50": 0.01993, "precision_at_100": 0.01, "mrr_at_1": 0.7785714285714286, "mrr_at_3": 0.8351190476190476, "mrr_at_5": 0.8429761904761904, "mrr_at_10": 0.8459807256235828, "mrr_at_20": 0.8478239927887484, "mrr_at_50": 0.8479991465823232, "mrr_at_100": 0.8480652841484608, "naucs_at_1_max": 0.5082211641432032, "naucs_at_1_std": 0.26902482824521823, "naucs_at_1_diff1": 0.8378750185346884, "naucs_at_3_max": 0.6534910260400478, "naucs_at_3_std": 0.5397344122834327, "naucs_at_3_diff1": 0.8195698032299361, "naucs_at_5_max": 0.6372274399956075, "naucs_at_5_std": 0.5036249794035255, "naucs_at_5_diff1": 0.8168286922612179, "naucs_at_10_max": 0.5800016976487535, "naucs_at_10_std": 0.4681266445972378, "naucs_at_10_diff1": 0.8242933537051196, "naucs_at_20_max": 0.478835978836005, "naucs_at_20_std": 0.08667911609088395, "naucs_at_20_diff1": 0.807812013694365, "naucs_at_50_max": 1.0, "naucs_at_50_std": 1.0, "naucs_at_50_diff1": 0.8692810457515607, "naucs_at_100_max": 1.0, "naucs_at_100_std": 1.0, "naucs_at_100_diff1": 1.0}, "tatdqa": {"ndcg_at_1": 0.54841, "ndcg_at_3": 0.6562, "ndcg_at_5": 0.68326, "ndcg_at_10": 0.70893, "ndcg_at_20": 0.72222, "ndcg_at_50": 0.7284, "ndcg_at_100": 0.73201, "map_at_1": 0.54841, "map_at_3": 0.63039, "map_at_5": 0.6453, "map_at_10": 0.65601, "map_at_20": 0.65978, "map_at_50": 0.6608, "map_at_100": 0.66112, "recall_at_1": 0.54841, "recall_at_3": 0.73061, "recall_at_5": 0.79675, "recall_at_10": 0.87553, "recall_at_20": 0.92724, "recall_at_50": 0.95791, "recall_at_100": 0.98016, "precision_at_1": 0.54841, "precision_at_3": 0.24354, "precision_at_5": 0.15935, "precision_at_10": 0.08755, "precision_at_20": 0.04636, "precision_at_50": 0.01916, "precision_at_100": 0.0098, "mrr_at_1": 0.5453998797354179, "mrr_at_3": 0.6282822208859491, "mrr_at_5": 0.6424133092804168, "mrr_at_10": 0.6538699424448072, "mrr_at_20": 0.657559818963894, "mrr_at_50": 0.6584837479678501, "mrr_at_100": 0.6587944596871749, "naucs_at_1_max": 0.3090653897821844, "naucs_at_1_std": 0.04930923898225459, "naucs_at_1_diff1": 0.7043054798202365, "naucs_at_3_max": 0.3130488042819241, "naucs_at_3_std": 0.10722995191204956, "naucs_at_3_diff1": 0.6136469206170786, "naucs_at_5_max": 0.3106346856808471, "naucs_at_5_std": 0.14652818200482404, "naucs_at_5_diff1": 0.5986106446949901, "naucs_at_10_max": 0.40632427830908485, "naucs_at_10_std": 0.29243794470603673, "naucs_at_10_diff1": 0.619230995521158, "naucs_at_20_max": 0.5480857947858365, "naucs_at_20_std": 0.47385204289648464, "naucs_at_20_diff1": 0.5875153603254067, "naucs_at_50_max": 0.6165541741820008, "naucs_at_50_std": 0.6292040856038491, "naucs_at_50_diff1": 0.596167522127439, "naucs_at_100_max": 0.6907473482145399, "naucs_at_100_std": 0.7358520747031099, "naucs_at_100_diff1": 0.6689903536377996}, "shift_project": {"ndcg_at_1": 0.53, "ndcg_at_3": 0.67357, "ndcg_at_5": 0.7024, "ndcg_at_10": 0.72124, "ndcg_at_20": 0.73433, "ndcg_at_50": 0.74196, "ndcg_at_100": 0.74196, "map_at_1": 0.53, "map_at_3": 0.64, "map_at_5": 0.656, "map_at_10": 0.66346, "map_at_20": 0.66728, "map_at_50": 0.66837, "map_at_100": 0.66837, "recall_at_1": 0.53, "recall_at_3": 0.77, "recall_at_5": 0.84, "recall_at_10": 0.9, "recall_at_20": 0.95, "recall_at_50": 0.99, "recall_at_100": 0.99, "precision_at_1": 0.53, "precision_at_3": 0.25667, "precision_at_5": 0.168, "precision_at_10": 0.09, "precision_at_20": 0.0475, "precision_at_50": 0.0198, "precision_at_100": 0.0099, "mrr_at_1": 0.54, "mrr_at_3": 0.6516666666666667, "mrr_at_5": 0.6651666666666667, "mrr_at_10": 0.6737341269841269, "mrr_at_20": 0.6768915945165945, "mrr_at_50": 0.6779969181064383, "mrr_at_100": 0.6779969181064383, "naucs_at_1_max": 0.04138888507058511, "naucs_at_1_std": -0.2798938817028416, "naucs_at_1_diff1": 0.7153362932823818, "naucs_at_3_max": 0.057585998522862565, "naucs_at_3_std": -0.34171124191490737, "naucs_at_3_diff1": 0.6187193661735401, "naucs_at_5_max": 0.3037248404516453, "naucs_at_5_std": -0.2680412371134025, "naucs_at_5_diff1": 0.6450662739322527, "naucs_at_10_max": 0.2270774976657352, "naucs_at_10_std": -0.29145658263305096, "naucs_at_10_diff1": 0.5507469654528476, "naucs_at_20_max": -0.005415499533139113, "naucs_at_20_std": -0.2233426704014865, "naucs_at_20_diff1": 0.32791783380019096, "naucs_at_50_max": 0.554154995331464, "naucs_at_50_std": 0.35807656395892007, "naucs_at_50_diff1": 0.12278244631185525, "naucs_at_100_max": 0.554154995331464, "naucs_at_100_std": 0.35807656395892007, "naucs_at_100_diff1": 0.12278244631185525}}
+{"./data_dir/eval_vidore/arxivqa_test_subsampled": {"ndcg_at_1": 0.766, "ndcg_at_3": 0.80828, "ndcg_at_5": 0.82464, "ndcg_at_10": 0.83726, "ndcg_at_20": 0.84494, "ndcg_at_50": 0.85016, "ndcg_at_100": 0.85311, "map_at_1": 0.766, "map_at_3": 0.798, "map_at_5": 0.807, "map_at_10": 0.81241, "map_at_20": 0.81457, "map_at_50": 0.81542, "map_at_100": 0.8157, "recall_at_1": 0.766, "recall_at_3": 0.838, "recall_at_5": 0.878, "recall_at_10": 0.916, "recall_at_20": 0.946, "recall_at_50": 0.972, "recall_at_100": 0.99, "precision_at_1": 0.766, "precision_at_3": 0.27933, "precision_at_5": 0.1756, "precision_at_10": 0.0916, "precision_at_20": 0.0473, "precision_at_50": 0.01944, "precision_at_100": 0.0099, "mrr_at_1": 0.764, "mrr_at_3": 0.7953333333333333, "mrr_at_5": 0.8052333333333334, "mrr_at_10": 0.8103563492063492, "mrr_at_20": 0.8125617465867466, "mrr_at_50": 0.8134103358953779, "mrr_at_100": 0.813679499264441, "naucs_at_1_max": 0.7002679923531673, "naucs_at_1_std": 0.04881257753897388, "naucs_at_1_diff1": 0.9187237029744059, "naucs_at_3_max": 0.7187632718933161, "naucs_at_3_std": 0.08625668903423199, "naucs_at_3_diff1": 0.898708910218297, "naucs_at_5_max": 0.7593782822028189, "naucs_at_5_std": 0.09854054647207043, "naucs_at_5_diff1": 0.8881111755576797, "naucs_at_10_max": 0.7248343781957253, "naucs_at_10_std": 0.038304210573119564, "naucs_at_10_diff1": 0.876061535725402, "naucs_at_20_max": 0.7865961199294541, "naucs_at_20_std": 0.07099629975446664, "naucs_at_20_diff1": 0.8862260953764205, "naucs_at_50_max": 0.8911231159130302, "naucs_at_50_std": 0.06549286381219031, "naucs_at_50_diff1": 0.8856209150326733, "naucs_at_100_max": 0.9183006535947666, "naucs_at_100_std": 0.05760971055089766, "naucs_at_100_diff1": 0.947712418300658}, "./data_dir/eval_vidore/docvqa_test_subsampled": {"ndcg_at_1": 0.51441, "ndcg_at_3": 0.57161, "ndcg_at_5": 0.58974, "ndcg_at_10": 0.6081, "ndcg_at_20": 0.62305, "ndcg_at_50": 0.6315, "ndcg_at_100": 0.63866, "map_at_1": 0.51441, "map_at_3": 0.55765, "map_at_5": 0.56763, "map_at_10": 0.57546, "map_at_20": 0.57947, "map_at_50": 0.58086, "map_at_100": 0.58148, "recall_at_1": 0.51441, "recall_at_3": 0.61197, "recall_at_5": 0.65632, "recall_at_10": 0.71175, "recall_at_20": 0.77162, "recall_at_50": 0.81375, "recall_at_100": 0.85809, "precision_at_1": 0.51441, "precision_at_3": 0.20399, "precision_at_5": 0.13126, "precision_at_10": 0.07118, "precision_at_20": 0.03858, "precision_at_50": 0.01627, "precision_at_100": 0.00858, "mrr_at_1": 0.5121951219512195, "mrr_at_3": 0.557280118255728, "mrr_at_5": 0.5679231337767923, "mrr_at_10": 0.5744324780910147, "mrr_at_20": 0.578282042051409, "mrr_at_50": 0.5799509163614124, "mrr_at_100": 0.5804699119047058, "naucs_at_1_max": 0.41316786027151814, "naucs_at_1_std": 0.6546349185814383, "naucs_at_1_diff1": 0.8366751779691564, "naucs_at_3_max": 0.3495069630602344, "naucs_at_3_std": 0.7551813316955129, "naucs_at_3_diff1": 0.7939263830700716, "naucs_at_5_max": 0.2979395852204152, "naucs_at_5_std": 0.8139018052278215, "naucs_at_5_diff1": 0.7690471628362231, "naucs_at_10_max": 0.21129358614554838, "naucs_at_10_std": 0.8562408620216512, "naucs_at_10_diff1": 0.7836510491949582, "naucs_at_20_max": 0.11661742091387217, "naucs_at_20_std": 0.8693119305170596, "naucs_at_20_diff1": 0.8000821971304247, "naucs_at_50_max": 0.0955474793604789, "naucs_at_50_std": 0.8573737499542009, "naucs_at_50_diff1": 0.7890856228265077, "naucs_at_100_max": 0.033291606120240604, "naucs_at_100_std": 0.8551707816577783, "naucs_at_100_diff1": 0.7966971634575577}, "./data_dir/eval_vidore/syntheticDocQA_energy_test": {"ndcg_at_1": 0.94, "ndcg_at_3": 0.95262, "ndcg_at_5": 0.95262, "ndcg_at_10": 0.95893, "ndcg_at_20": 0.95893, "ndcg_at_50": 0.96081, "ndcg_at_100": 0.96241, "map_at_1": 0.94, "map_at_3": 0.95, "map_at_5": 0.95, "map_at_10": 0.9525, "map_at_20": 0.9525, "map_at_50": 0.95276, "map_at_100": 0.95289, "recall_at_1": 0.94, "recall_at_3": 0.96, "recall_at_5": 0.96, "recall_at_10": 0.98, "recall_at_20": 0.98, "recall_at_50": 0.99, "recall_at_100": 1.0, "precision_at_1": 0.94, "precision_at_3": 0.32, "precision_at_5": 0.192, "precision_at_10": 0.098, "precision_at_20": 0.049, "precision_at_50": 0.0198, "precision_at_100": 0.01, "mrr_at_1": 0.95, "mrr_at_3": 0.955, "mrr_at_5": 0.955, "mrr_at_10": 0.9575, "mrr_at_20": 0.9575, "mrr_at_50": 0.9577564102564102, "mrr_at_100": 0.95789339655778, "naucs_at_1_max": 0.3629007158418935, "naucs_at_1_std": -0.6582633053221264, "naucs_at_1_diff1": 0.9782135076252712, "naucs_at_3_max": 0.44864612511671187, "naucs_at_3_std": -0.6419234360410808, "naucs_at_3_diff1": 0.9673202614379101, "naucs_at_5_max": 0.44864612511671176, "naucs_at_5_std": -0.6419234360410755, "naucs_at_5_diff1": 0.967320261437907, "naucs_at_10_max": 1.0, "naucs_at_10_std": 0.45611577964519334, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 1.0, "naucs_at_20_std": 0.45611577964519334, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": 1.0, "naucs_at_50_std": 0.554154995331464, "naucs_at_50_diff1": 1.0, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/tatdqa_test": {"ndcg_at_1": 0.55589, "ndcg_at_3": 0.65956, "ndcg_at_5": 0.68639, "ndcg_at_10": 0.71121, "ndcg_at_20": 0.72342, "ndcg_at_50": 0.73233, "ndcg_at_100": 0.7349, "map_at_1": 0.55589, "map_at_3": 0.63426, "map_at_5": 0.64918, "map_at_10": 0.65958, "map_at_20": 0.6629, "map_at_50": 0.66442, "map_at_100": 0.66465, "recall_at_1": 0.55589, "recall_at_3": 0.73269, "recall_at_5": 0.79769, "recall_at_10": 0.87363, "recall_at_20": 0.92224, "recall_at_50": 0.96598, "recall_at_100": 0.98177, "precision_at_1": 0.55589, "precision_at_3": 0.24423, "precision_at_5": 0.15954, "precision_at_10": 0.08736, "precision_at_20": 0.04611, "precision_at_50": 0.01932, "precision_at_100": 0.00982, "mrr_at_1": 0.554678007290401, "mrr_at_3": 0.6337586067233698, "mrr_at_5": 0.6497367355204536, "mrr_at_10": 0.6598222723678373, "mrr_at_20": 0.6631270222189808, "mrr_at_50": 0.6646612644237917, "mrr_at_100": 0.664891463227161, "naucs_at_1_max": 0.23432149786481715, "naucs_at_1_std": -0.060743790148545855, "naucs_at_1_diff1": 0.685195965555728, "naucs_at_3_max": 0.28181099513082436, "naucs_at_3_std": 0.010556836141641941, "naucs_at_3_diff1": 0.572008448605137, "naucs_at_5_max": 0.32064142779819044, "naucs_at_5_std": 0.10890192999138859, "naucs_at_5_diff1": 0.5661052532485314, "naucs_at_10_max": 0.3583128977278265, "naucs_at_10_std": 0.21671540449890933, "naucs_at_10_diff1": 0.5371225125285051, "naucs_at_20_max": 0.4004094077799751, "naucs_at_20_std": 0.3159983579150967, "naucs_at_20_diff1": 0.5104165269228296, "naucs_at_50_max": 0.5437534661131097, "naucs_at_50_std": 0.4908038661588003, "naucs_at_50_diff1": 0.6100780162875041, "naucs_at_100_max": 0.5230675889370707, "naucs_at_100_std": 0.5481653625823034, "naucs_at_100_diff1": 0.6338872974520477}, "./data_dir/eval_vidore/infovqa_test_subsampled": {"ndcg_at_1": 0.77733, "ndcg_at_3": 0.82907, "ndcg_at_5": 0.84152, "ndcg_at_10": 0.85359, "ndcg_at_20": 0.85926, "ndcg_at_50": 0.86163, "ndcg_at_100": 0.86456, "map_at_1": 0.77733, "map_at_3": 0.8168, "map_at_5": 0.82368, "map_at_10": 0.82845, "map_at_20": 0.83002, "map_at_50": 0.83039, "map_at_100": 0.83064, "recall_at_1": 0.77733, "recall_at_3": 0.86437, "recall_at_5": 0.89474, "recall_at_10": 0.9332, "recall_at_20": 0.95547, "recall_at_50": 0.96761, "recall_at_100": 0.98583, "precision_at_1": 0.77733, "precision_at_3": 0.28812, "precision_at_5": 0.17895, "precision_at_10": 0.09332, "precision_at_20": 0.04777, "precision_at_50": 0.01935, "precision_at_100": 0.00986, "mrr_at_1": 0.7793522267206477, "mrr_at_3": 0.8191632928475034, "mrr_at_5": 0.8249325236167341, "mrr_at_10": 0.8297450356660883, "mrr_at_20": 0.8313321368003874, "mrr_at_50": 0.831710585121861, "mrr_at_100": 0.8319356459461824, "naucs_at_1_max": 0.6207806057689589, "naucs_at_1_std": 0.09455453314004965, "naucs_at_1_diff1": 0.9179386604371041, "naucs_at_3_max": 0.7135513762466027, "naucs_at_3_std": 0.20795232669207667, "naucs_at_3_diff1": 0.8822792301659564, "naucs_at_5_max": 0.6873058734577542, "naucs_at_5_std": 0.20627498312628267, "naucs_at_5_diff1": 0.8552616332944558, "naucs_at_10_max": 0.8492458496569022, "naucs_at_10_std": 0.3994979919194544, "naucs_at_10_diff1": 0.8329890886766317, "naucs_at_20_max": 0.8852384363992515, "naucs_at_20_std": 0.5174153630934645, "naucs_at_20_diff1": 0.8170387004512121, "naucs_at_50_max": 0.870059401046633, "naucs_at_50_std": 0.5160860129547604, "naucs_at_50_diff1": 0.7994702370982573, "naucs_at_100_max": 0.9043515571496571, "naucs_at_100_std": 0.6212420507953942, "naucs_at_100_diff1": 0.7886346576649294}, "./data_dir/eval_vidore/syntheticDocQA_healthcare_industry_test": {"ndcg_at_1": 0.93, "ndcg_at_3": 0.96786, "ndcg_at_5": 0.97216, "ndcg_at_10": 0.97216, "ndcg_at_20": 0.97216, "ndcg_at_50": 0.97216, "ndcg_at_100": 0.97216, "map_at_1": 0.93, "map_at_3": 0.96, "map_at_5": 0.9625, "map_at_10": 0.9625, "map_at_20": 0.9625, "map_at_50": 0.9625, "map_at_100": 0.9625, "recall_at_1": 0.93, "recall_at_3": 0.99, "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": 0.33, "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": 0.965, "mrr_at_5": 0.9675, "mrr_at_10": 0.9675, "mrr_at_20": 0.9675, "mrr_at_50": 0.9675, "mrr_at_100": 0.9675, "naucs_at_1_max": 0.17146858743497356, "naucs_at_1_std": -0.7483660130718978, "naucs_at_1_diff1": 0.8569427771108457, "naucs_at_3_max": -0.5634920634921204, "naucs_at_3_std": -1.7399626517274398, "naucs_at_3_diff1": 0.5541549953314585, "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/tabfquad_test_subsampled": {"ndcg_at_1": 0.825, "ndcg_at_3": 0.89064, "ndcg_at_5": 0.9054, "ndcg_at_10": 0.91102, "ndcg_at_20": 0.91272, "ndcg_at_50": 0.91345, "ndcg_at_100": 0.91345, "map_at_1": 0.825, "map_at_3": 0.875, "map_at_5": 0.88321, "map_at_10": 0.88545, "map_at_20": 0.88585, "map_at_50": 0.88598, "map_at_100": 0.88598, "recall_at_1": 0.825, "recall_at_3": 0.93571, "recall_at_5": 0.97143, "recall_at_10": 0.98929, "recall_at_20": 0.99643, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.825, "precision_at_3": 0.3119, "precision_at_5": 0.19429, "precision_at_10": 0.09893, "precision_at_20": 0.04982, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.8285714285714286, "mrr_at_3": 0.8761904761904762, "mrr_at_5": 0.8845833333333334, "mrr_at_10": 0.8870280612244897, "mrr_at_20": 0.887436557956516, "mrr_at_50": 0.8875597106658755, "mrr_at_100": 0.8875597106658755, "naucs_at_1_max": 0.5575808958598282, "naucs_at_1_std": 0.2596035446819686, "naucs_at_1_diff1": 0.8951432204929446, "naucs_at_3_max": 0.61155202821869, "naucs_at_3_std": 0.4423436041083083, "naucs_at_3_diff1": 0.8163969291420269, "naucs_at_5_max": 0.7310340802987914, "naucs_at_5_std": 0.6248249299719922, "naucs_at_5_diff1": 0.8421451914098955, "naucs_at_10_max": 0.28275754746343684, "naucs_at_10_std": 0.043106131341434525, "naucs_at_10_diff1": 0.8148148148148323, "naucs_at_20_max": 1.0, "naucs_at_20_std": 0.8692810457516478, "naucs_at_20_diff1": 1.0, "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}, "./data_dir/eval_vidore/syntheticDocQA_government_reports_test": {"ndcg_at_1": 0.86, "ndcg_at_3": 0.94071, "ndcg_at_5": 0.94071, "ndcg_at_10": 0.94404, "ndcg_at_20": 0.94404, "ndcg_at_50": 0.94404, "ndcg_at_100": 0.94404, "map_at_1": 0.86, "map_at_3": 0.92333, "map_at_5": 0.92333, "map_at_10": 0.92476, "map_at_20": 0.92476, "map_at_50": 0.92476, "map_at_100": 0.92476, "recall_at_1": 0.86, "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": 0.86, "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": 0.86, "mrr_at_3": 0.9233333333333333, "mrr_at_5": 0.9233333333333333, "mrr_at_10": 0.9247619047619048, "mrr_at_20": 0.9247619047619048, "mrr_at_50": 0.9247619047619048, "mrr_at_100": 0.9247619047619048, "naucs_at_1_max": 0.052653457115144964, "naucs_at_1_std": -0.5728834695857007, "naucs_at_1_diff1": 0.9159969516419558, "naucs_at_3_max": 1.0, "naucs_at_3_std": 0.5541549953314585, "naucs_at_3_diff1": 0.8692810457516356, "naucs_at_5_max": 1.0, "naucs_at_5_std": 0.5541549953314738, "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}, "./data_dir/eval_vidore/shiftproject_test": {"ndcg_at_1": 0.6, "ndcg_at_3": 0.74202, "ndcg_at_5": 0.75881, "ndcg_at_10": 0.76571, "ndcg_at_20": 0.77364, "ndcg_at_50": 0.78175, "ndcg_at_100": 0.78492, "map_at_1": 0.6, "map_at_3": 0.705, "map_at_5": 0.7145, "map_at_10": 0.7176, "map_at_20": 0.71997, "map_at_50": 0.72135, "map_at_100": 0.7216, "recall_at_1": 0.6, "recall_at_3": 0.85, "recall_at_5": 0.89, "recall_at_10": 0.91, "recall_at_20": 0.94, "recall_at_50": 0.98, "recall_at_100": 1.0, "precision_at_1": 0.6, "precision_at_3": 0.28333, "precision_at_5": 0.178, "precision_at_10": 0.091, "precision_at_20": 0.047, "precision_at_50": 0.0196, "precision_at_100": 0.01, "mrr_at_1": 0.6, "mrr_at_3": 0.7083333333333333, "mrr_at_5": 0.7178333333333333, "mrr_at_10": 0.7209285714285714, "mrr_at_20": 0.7238717532467532, "mrr_at_50": 0.7248075998075998, "mrr_at_100": 0.7250711683874507, "naucs_at_1_max": 0.0373849807064409, "naucs_at_1_std": -0.1642030276046306, "naucs_at_1_diff1": 0.6065004452359747, "naucs_at_3_max": -0.000748454279205037, "naucs_at_3_std": -0.30231044581841743, "naucs_at_3_diff1": 0.628115847705825, "naucs_at_5_max": 0.14672279013830405, "naucs_at_5_std": -0.17313804655957335, "naucs_at_5_diff1": 0.6169143544369039, "naucs_at_10_max": 0.04082373690216766, "naucs_at_10_std": -0.2267351384998422, "naucs_at_10_diff1": 0.6086730988691776, "naucs_at_20_max": -0.10387488328664873, "naucs_at_20_std": -0.22829131652661377, "naucs_at_20_diff1": 0.6160908807967658, "naucs_at_50_max": 0.638188608776847, "naucs_at_50_std": -0.004668534080297299, "naucs_at_50_diff1": 0.7117180205415448, "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": 0.95, "ndcg_at_3": 0.97524, "ndcg_at_5": 0.97954, "ndcg_at_10": 0.97954, "ndcg_at_20": 0.97954, "ndcg_at_50": 0.97954, "ndcg_at_100": 0.97954, "map_at_1": 0.95, "map_at_3": 0.97, "map_at_5": 0.9725, "map_at_10": 0.9725, "map_at_20": 0.9725, "map_at_50": 0.9725, "map_at_100": 0.9725, "recall_at_1": 0.95, "recall_at_3": 0.99, "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": 0.33, "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": 0.95, "mrr_at_3": 0.97, "mrr_at_5": 0.9725, "mrr_at_10": 0.9725, "mrr_at_20": 0.9725, "mrr_at_50": 0.9725, "mrr_at_100": 0.9725, "naucs_at_1_max": -0.16125116713351925, "naucs_at_1_std": -0.20662931839401985, "naucs_at_1_diff1": 0.9183006535947691, "naucs_at_3_max": -1.1517273576097802, "naucs_at_3_std": -1.7399626517274398, "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}}