| { | |
| "vidore/arxivqa_test_subsampled": { | |
| "ndcg_at_1": 0.856, | |
| "ndcg_at_3": 0.90155, | |
| "ndcg_at_5": 0.90705, | |
| "ndcg_at_10": 0.91307, | |
| "ndcg_at_20": 0.91718, | |
| "ndcg_at_50": 0.91879, | |
| "ndcg_at_100": 0.91978, | |
| "map_at_1": 0.856, | |
| "map_at_3": 0.89033, | |
| "map_at_5": 0.89323, | |
| "map_at_10": 0.89584, | |
| "map_at_20": 0.897, | |
| "map_at_50": 0.89727, | |
| "map_at_100": 0.89736, | |
| "recall_at_1": 0.856, | |
| "recall_at_3": 0.934, | |
| "recall_at_5": 0.948, | |
| "recall_at_10": 0.966, | |
| "recall_at_20": 0.982, | |
| "recall_at_50": 0.99, | |
| "recall_at_100": 0.996, | |
| "precision_at_1": 0.856, | |
| "precision_at_3": 0.31133, | |
| "precision_at_5": 0.1896, | |
| "precision_at_10": 0.0966, | |
| "precision_at_20": 0.0491, | |
| "precision_at_50": 0.0198, | |
| "precision_at_100": 0.00996, | |
| "mrr_at_1": 0.856, | |
| "mrr_at_3": 0.8903333333333333, | |
| "mrr_at_5": 0.8937333333333332, | |
| "mrr_at_10": 0.8958880952380952, | |
| "mrr_at_20": 0.8972350796262559, | |
| "mrr_at_50": 0.8974958203669966, | |
| "mrr_at_100": 0.8975858521087404, | |
| "naucs_at_1_max": 0.7091055219364591, | |
| "naucs_at_1_std": -0.1539604495353359, | |
| "naucs_at_1_diff1": 0.9332045601901878, | |
| "naucs_at_3_max": 0.7513934866876091, | |
| "naucs_at_3_std": -0.0981099510511257, | |
| "naucs_at_3_diff1": 0.9176640353110959, | |
| "naucs_at_5_max": 0.7900237017884059, | |
| "naucs_at_5_std": -0.0631688572865076, | |
| "naucs_at_5_diff1": 0.9162357250592552, | |
| "naucs_at_10_max": 0.8500851320920556, | |
| "naucs_at_10_std": -0.01117701982753666, | |
| "naucs_at_10_diff1": 0.936562860438295, | |
| "naucs_at_20_max": 0.9709513435003626, | |
| "naucs_at_20_std": 0.23140367258011904, | |
| "naucs_at_20_diff1": 0.9546114742193182, | |
| "naucs_at_50_max": 0.9738562091503188, | |
| "naucs_at_50_std": 0.5435107376283843, | |
| "naucs_at_50_diff1": 0.9183006535947666, | |
| "naucs_at_100_max": 0.9346405228758466, | |
| "naucs_at_100_std": -0.07586367880485818, | |
| "naucs_at_100_diff1": 1.0 | |
| }, | |
| "vidore/docvqa_test_subsampled": { | |
| "ndcg_at_1": 0.54324, | |
| "ndcg_at_3": 0.60661, | |
| "ndcg_at_5": 0.63199, | |
| "ndcg_at_10": 0.65541, | |
| "ndcg_at_20": 0.66441, | |
| "ndcg_at_50": 0.6786, | |
| "ndcg_at_100": 0.68571, | |
| "map_at_1": 0.54324, | |
| "map_at_3": 0.59165, | |
| "map_at_5": 0.60562, | |
| "map_at_10": 0.61514, | |
| "map_at_20": 0.61762, | |
| "map_at_50": 0.61995, | |
| "map_at_100": 0.62055, | |
| "recall_at_1": 0.54324, | |
| "recall_at_3": 0.64967, | |
| "recall_at_5": 0.71175, | |
| "recall_at_10": 0.78492, | |
| "recall_at_20": 0.8204, | |
| "recall_at_50": 0.89135, | |
| "recall_at_100": 0.9357, | |
| "precision_at_1": 0.54324, | |
| "precision_at_3": 0.21656, | |
| "precision_at_5": 0.14235, | |
| "precision_at_10": 0.07849, | |
| "precision_at_20": 0.04102, | |
| "precision_at_50": 0.01783, | |
| "precision_at_100": 0.00936, | |
| "mrr_at_1": 0.541019955654102, | |
| "mrr_at_3": 0.5923872875092389, | |
| "mrr_at_5": 0.6049150036954914, | |
| "mrr_at_10": 0.6141546123253439, | |
| "mrr_at_20": 0.6172553927167855, | |
| "mrr_at_50": 0.619590892958523, | |
| "mrr_at_100": 0.6201350422138927, | |
| "naucs_at_1_max": 0.6070360156876232, | |
| "naucs_at_1_std": -0.291121623963555, | |
| "naucs_at_1_diff1": 0.8758394294393667, | |
| "naucs_at_3_max": 0.5616540550792796, | |
| "naucs_at_3_std": -0.3488346790402726, | |
| "naucs_at_3_diff1": 0.8249830639439742, | |
| "naucs_at_5_max": 0.5498221632518324, | |
| "naucs_at_5_std": -0.3185102580355566, | |
| "naucs_at_5_diff1": 0.7690366497607697, | |
| "naucs_at_10_max": 0.5941928005860726, | |
| "naucs_at_10_std": -0.19752553123275665, | |
| "naucs_at_10_diff1": 0.7514730263777738, | |
| "naucs_at_20_max": 0.5661659566373964, | |
| "naucs_at_20_std": -0.2142692317658413, | |
| "naucs_at_20_diff1": 0.7481778263570004, | |
| "naucs_at_50_max": 0.5398055368355262, | |
| "naucs_at_50_std": 0.025022300394388387, | |
| "naucs_at_50_diff1": 0.7040325634625932, | |
| "naucs_at_100_max": 0.5303081878412225, | |
| "naucs_at_100_std": 0.1831093237519082, | |
| "naucs_at_100_diff1": 0.6786345214154387 | |
| }, | |
| "vidore/infovqa_test_subsampled": { | |
| "ndcg_at_1": 0.90283, | |
| "ndcg_at_3": 0.92753, | |
| "ndcg_at_5": 0.9318, | |
| "ndcg_at_10": 0.93516, | |
| "ndcg_at_20": 0.93777, | |
| "ndcg_at_50": 0.93855, | |
| "ndcg_at_100": 0.94024, | |
| "map_at_1": 0.90283, | |
| "map_at_3": 0.92139, | |
| "map_at_5": 0.92382, | |
| "map_at_10": 0.92526, | |
| "map_at_20": 0.926, | |
| "map_at_50": 0.92611, | |
| "map_at_100": 0.92628, | |
| "recall_at_1": 0.90283, | |
| "recall_at_3": 0.94534, | |
| "recall_at_5": 0.95547, | |
| "recall_at_10": 0.96559, | |
| "recall_at_20": 0.97571, | |
| "recall_at_50": 0.97976, | |
| "recall_at_100": 0.98988, | |
| "precision_at_1": 0.90283, | |
| "precision_at_3": 0.31511, | |
| "precision_at_5": 0.19109, | |
| "precision_at_10": 0.09656, | |
| "precision_at_20": 0.04879, | |
| "precision_at_50": 0.0196, | |
| "precision_at_100": 0.0099, | |
| "mrr_at_1": 0.902834008097166, | |
| "mrr_at_3": 0.9210526315789471, | |
| "mrr_at_5": 0.9234817813765179, | |
| "mrr_at_10": 0.9249212775528565, | |
| "mrr_at_20": 0.9256530244384494, | |
| "mrr_at_50": 0.9257658332525782, | |
| "mrr_at_100": 0.9259269979005336, | |
| "naucs_at_1_max": 0.7169019957672044, | |
| "naucs_at_1_std": 0.0348078867315928, | |
| "naucs_at_1_diff1": 0.932198694378402, | |
| "naucs_at_3_max": 0.6993822189004274, | |
| "naucs_at_3_std": -0.05440424958192625, | |
| "naucs_at_3_diff1": 0.9510168011008809, | |
| "naucs_at_5_max": 0.8005964299196806, | |
| "naucs_at_5_std": 0.020473977525333666, | |
| "naucs_at_5_diff1": 0.939884255896541, | |
| "naucs_at_10_max": 0.8726100984749249, | |
| "naucs_at_10_std": 0.20123744718206102, | |
| "naucs_at_10_diff1": 0.9375680316384316, | |
| "naucs_at_20_max": 0.9193241139815224, | |
| "naucs_at_20_std": 0.3738448936124196, | |
| "naucs_at_20_diff1": 0.9333216204985769, | |
| "naucs_at_50_max": 0.960819563780575, | |
| "naucs_at_50_std": 0.5807604079447772, | |
| "naucs_at_50_diff1": 0.9477594183740831, | |
| "naucs_at_100_max": 0.9477594183740937, | |
| "naucs_at_100_std": 0.7984112420199415, | |
| "naucs_at_100_diff1": 0.9216391275611305 | |
| }, | |
| "vidore/tabfquad_test_subsampled": { | |
| "ndcg_at_1": 0.91786, | |
| "ndcg_at_3": 0.94979, | |
| "ndcg_at_5": 0.95117, | |
| "ndcg_at_10": 0.9546, | |
| "ndcg_at_20": 0.95812, | |
| "ndcg_at_50": 0.95812, | |
| "ndcg_at_100": 0.95812, | |
| "map_at_1": 0.91786, | |
| "map_at_3": 0.94226, | |
| "map_at_5": 0.94298, | |
| "map_at_10": 0.94438, | |
| "map_at_20": 0.9453, | |
| "map_at_50": 0.9453, | |
| "map_at_100": 0.9453, | |
| "recall_at_1": 0.91786, | |
| "recall_at_3": 0.97143, | |
| "recall_at_5": 0.975, | |
| "recall_at_10": 0.98571, | |
| "recall_at_20": 1.0, | |
| "recall_at_50": 1.0, | |
| "recall_at_100": 1.0, | |
| "precision_at_1": 0.91786, | |
| "precision_at_3": 0.32381, | |
| "precision_at_5": 0.195, | |
| "precision_at_10": 0.09857, | |
| "precision_at_20": 0.05, | |
| "precision_at_50": 0.02, | |
| "precision_at_100": 0.01, | |
| "mrr_at_1": 0.9107142857142857, | |
| "mrr_at_3": 0.9386904761904761, | |
| "mrr_at_5": 0.939404761904762, | |
| "mrr_at_10": 0.9408035714285715, | |
| "mrr_at_20": 0.9417180023923446, | |
| "mrr_at_50": 0.9417180023923446, | |
| "mrr_at_100": 0.9417180023923446, | |
| "naucs_at_1_max": 0.673689765761377, | |
| "naucs_at_1_std": 0.22151179312304817, | |
| "naucs_at_1_diff1": 0.9465148378191858, | |
| "naucs_at_3_max": 0.9489379084967345, | |
| "naucs_at_3_std": 0.7698412698412723, | |
| "naucs_at_3_diff1": 1.0, | |
| "naucs_at_5_max": 0.9416433239962654, | |
| "naucs_at_5_std": 0.7766439909297053, | |
| "naucs_at_5_diff1": 1.0, | |
| "naucs_at_10_max": 0.9305555555555624, | |
| "naucs_at_10_std": 0.7112511671335298, | |
| "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": 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 | |
| }, | |
| "vidore/tatdqa_test": { | |
| "ndcg_at_1": 0.68894, | |
| "ndcg_at_3": 0.77901, | |
| "ndcg_at_5": 0.80166, | |
| "ndcg_at_10": 0.81543, | |
| "ndcg_at_20": 0.8202, | |
| "ndcg_at_50": 0.82432, | |
| "ndcg_at_100": 0.8266, | |
| "map_at_1": 0.68894, | |
| "map_at_3": 0.75719, | |
| "map_at_5": 0.76983, | |
| "map_at_10": 0.77563, | |
| "map_at_20": 0.77694, | |
| "map_at_50": 0.77761, | |
| "map_at_100": 0.77782, | |
| "recall_at_1": 0.68894, | |
| "recall_at_3": 0.84204, | |
| "recall_at_5": 0.89672, | |
| "recall_at_10": 0.93864, | |
| "recall_at_20": 0.95747, | |
| "recall_at_50": 0.97813, | |
| "recall_at_100": 0.9921, | |
| "precision_at_1": 0.68894, | |
| "precision_at_3": 0.28068, | |
| "precision_at_5": 0.17934, | |
| "precision_at_10": 0.09386, | |
| "precision_at_20": 0.04787, | |
| "precision_at_50": 0.01956, | |
| "precision_at_100": 0.00992, | |
| "mrr_at_1": 0.68955042527339, | |
| "mrr_at_3": 0.7578979343863913, | |
| "mrr_at_5": 0.7698967193195627, | |
| "mrr_at_10": 0.7759968851858281, | |
| "mrr_at_20": 0.7772993998806929, | |
| "mrr_at_50": 0.777976076361586, | |
| "mrr_at_100": 0.7781777477659721, | |
| "naucs_at_1_max": 0.18182325832334814, | |
| "naucs_at_1_std": -0.2986720581428371, | |
| "naucs_at_1_diff1": 0.7996863187967407, | |
| "naucs_at_3_max": 0.20431138412136718, | |
| "naucs_at_3_std": -0.2642639510023209, | |
| "naucs_at_3_diff1": 0.7009272933336391, | |
| "naucs_at_5_max": 0.2908041510404104, | |
| "naucs_at_5_std": -0.1423448642637957, | |
| "naucs_at_5_diff1": 0.659321151493992, | |
| "naucs_at_10_max": 0.319161841315299, | |
| "naucs_at_10_std": 0.023933713165685305, | |
| "naucs_at_10_diff1": 0.6512910283423078, | |
| "naucs_at_20_max": 0.3535520101617024, | |
| "naucs_at_20_std": 0.12052219709226725, | |
| "naucs_at_20_diff1": 0.6423271539480547, | |
| "naucs_at_50_max": 0.3083707088493802, | |
| "naucs_at_50_std": 0.306461885213115, | |
| "naucs_at_50_diff1": 0.6344678950329469, | |
| "naucs_at_100_max": 0.6291544556839062, | |
| "naucs_at_100_std": 0.7759386182684255, | |
| "naucs_at_100_diff1": 0.6495298769820174 | |
| }, | |
| "vidore/shiftproject_test": { | |
| "ndcg_at_1": 0.74, | |
| "ndcg_at_3": 0.87857, | |
| "ndcg_at_5": 0.87857, | |
| "ndcg_at_10": 0.88528, | |
| "ndcg_at_20": 0.88528, | |
| "ndcg_at_50": 0.88715, | |
| "ndcg_at_100": 0.88715, | |
| "map_at_1": 0.74, | |
| "map_at_3": 0.84667, | |
| "map_at_5": 0.84667, | |
| "map_at_10": 0.84958, | |
| "map_at_20": 0.84958, | |
| "map_at_50": 0.84983, | |
| "map_at_100": 0.84983, | |
| "recall_at_1": 0.74, | |
| "recall_at_3": 0.97, | |
| "recall_at_5": 0.97, | |
| "recall_at_10": 0.99, | |
| "recall_at_20": 0.99, | |
| "recall_at_50": 1.0, | |
| "recall_at_100": 1.0, | |
| "precision_at_1": 0.74, | |
| "precision_at_3": 0.32333, | |
| "precision_at_5": 0.194, | |
| "precision_at_10": 0.099, | |
| "precision_at_20": 0.0495, | |
| "precision_at_50": 0.02, | |
| "precision_at_100": 0.01, | |
| "mrr_at_1": 0.76, | |
| "mrr_at_3": 0.8583333333333333, | |
| "mrr_at_5": 0.8603333333333335, | |
| "mrr_at_10": 0.8615833333333335, | |
| "mrr_at_20": 0.8615833333333335, | |
| "mrr_at_50": 0.8618333333333333, | |
| "mrr_at_100": 0.8618333333333333, | |
| "naucs_at_1_max": 0.23297919792384728, | |
| "naucs_at_1_std": -0.24611735128340276, | |
| "naucs_at_1_diff1": 0.8178500466323342, | |
| "naucs_at_3_max": 0.6374105197634555, | |
| "naucs_at_3_std": 0.5169623404917584, | |
| "naucs_at_3_diff1": 0.6008403361344542, | |
| "naucs_at_5_max": 0.6374105197634568, | |
| "naucs_at_5_std": 0.5169623404917556, | |
| "naucs_at_5_diff1": 0.6008403361344512, | |
| "naucs_at_10_max": 0.5541549953314738, | |
| "naucs_at_10_std": 0.7222222222222276, | |
| "naucs_at_10_diff1": 0.7222222222222276, | |
| "naucs_at_20_max": 0.5541549953314738, | |
| "naucs_at_20_std": 0.7222222222222276, | |
| "naucs_at_20_diff1": 0.7222222222222276, | |
| "naucs_at_50_max": null, | |
| "naucs_at_50_std": null, | |
| "naucs_at_50_diff1": null, | |
| "naucs_at_100_max": null, | |
| "naucs_at_100_std": null, | |
| "naucs_at_100_diff1": null | |
| }, | |
| "vidore/syntheticDocQA_artificial_intelligence_test": { | |
| "ndcg_at_1": 0.97, | |
| "ndcg_at_3": 0.98893, | |
| "ndcg_at_5": 0.98893, | |
| "ndcg_at_10": 0.98893, | |
| "ndcg_at_20": 0.98893, | |
| "ndcg_at_50": 0.98893, | |
| "ndcg_at_100": 0.98893, | |
| "map_at_1": 0.97, | |
| "map_at_3": 0.985, | |
| "map_at_5": 0.985, | |
| "map_at_10": 0.985, | |
| "map_at_20": 0.985, | |
| "map_at_50": 0.985, | |
| "map_at_100": 0.985, | |
| "recall_at_1": 0.97, | |
| "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": 0.97, | |
| "precision_at_3": 0.33333, | |
| "precision_at_5": 0.2, | |
| "precision_at_10": 0.1, | |
| "precision_at_20": 0.05, | |
| "precision_at_50": 0.02, | |
| "precision_at_100": 0.01, | |
| "mrr_at_1": 0.97, | |
| "mrr_at_3": 0.985, | |
| "mrr_at_5": 0.985, | |
| "mrr_at_10": 0.985, | |
| "mrr_at_20": 0.985, | |
| "mrr_at_50": 0.985, | |
| "mrr_at_100": 0.985, | |
| "naucs_at_1_max": 0.49361967009025914, | |
| "naucs_at_1_std": 0.266573295985059, | |
| "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": null, | |
| "naucs_at_50_std": null, | |
| "naucs_at_50_diff1": null, | |
| "naucs_at_100_max": null, | |
| "naucs_at_100_std": null, | |
| "naucs_at_100_diff1": null | |
| }, | |
| "vidore/syntheticDocQA_energy_test": { | |
| "ndcg_at_1": 0.95, | |
| "ndcg_at_3": 0.96262, | |
| "ndcg_at_5": 0.97079, | |
| "ndcg_at_10": 0.97079, | |
| "ndcg_at_20": 0.97079, | |
| "ndcg_at_50": 0.97281, | |
| "ndcg_at_100": 0.97281, | |
| "map_at_1": 0.95, | |
| "map_at_3": 0.96, | |
| "map_at_5": 0.9645, | |
| "map_at_10": 0.9645, | |
| "map_at_20": 0.9645, | |
| "map_at_50": 0.96483, | |
| "map_at_100": 0.96483, | |
| "recall_at_1": 0.95, | |
| "recall_at_3": 0.97, | |
| "recall_at_5": 0.99, | |
| "recall_at_10": 0.99, | |
| "recall_at_20": 0.99, | |
| "recall_at_50": 1.0, | |
| "recall_at_100": 1.0, | |
| "precision_at_1": 0.95, | |
| "precision_at_3": 0.32333, | |
| "precision_at_5": 0.198, | |
| "precision_at_10": 0.099, | |
| "precision_at_20": 0.0495, | |
| "precision_at_50": 0.02, | |
| "precision_at_100": 0.01, | |
| "mrr_at_1": 0.96, | |
| "mrr_at_3": 0.965, | |
| "mrr_at_5": 0.97, | |
| "mrr_at_10": 0.97, | |
| "mrr_at_20": 0.97, | |
| "mrr_at_50": 0.970344827586207, | |
| "mrr_at_100": 0.970344827586207, | |
| "naucs_at_1_max": 0.40961718020541377, | |
| "naucs_at_1_std": -1.1517273576097071, | |
| "naucs_at_1_diff1": 1.0, | |
| "naucs_at_3_max": 0.6988795518207244, | |
| "naucs_at_3_std": -0.9556489262371565, | |
| "naucs_at_3_diff1": 1.0, | |
| "naucs_at_5_max": 0.8692810457516413, | |
| "naucs_at_5_std": -0.5634920634920767, | |
| "naucs_at_5_diff1": 1.0, | |
| "naucs_at_10_max": 0.8692810457516413, | |
| "naucs_at_10_std": -0.5634920634920767, | |
| "naucs_at_10_diff1": 1.0, | |
| "naucs_at_20_max": 0.8692810457516413, | |
| "naucs_at_20_std": -0.5634920634920767, | |
| "naucs_at_20_diff1": 1.0, | |
| "naucs_at_50_max": null, | |
| "naucs_at_50_std": null, | |
| "naucs_at_50_diff1": null, | |
| "naucs_at_100_max": null, | |
| "naucs_at_100_std": null, | |
| "naucs_at_100_diff1": null | |
| }, | |
| "vidore/syntheticDocQA_government_reports_test": { | |
| "ndcg_at_1": 0.92, | |
| "ndcg_at_3": 0.95524, | |
| "ndcg_at_5": 0.95954, | |
| "ndcg_at_10": 0.96311, | |
| "ndcg_at_20": 0.96311, | |
| "ndcg_at_50": 0.96311, | |
| "ndcg_at_100": 0.96311, | |
| "map_at_1": 0.92, | |
| "map_at_3": 0.94667, | |
| "map_at_5": 0.94917, | |
| "map_at_10": 0.95083, | |
| "map_at_20": 0.95083, | |
| "map_at_50": 0.95083, | |
| "map_at_100": 0.95083, | |
| "recall_at_1": 0.92, | |
| "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": 0.92, | |
| "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": 0.92, | |
| "mrr_at_3": 0.9483333333333333, | |
| "mrr_at_5": 0.9508333333333333, | |
| "mrr_at_10": 0.9524999999999999, | |
| "mrr_at_20": 0.9524999999999999, | |
| "mrr_at_50": 0.9524999999999999, | |
| "mrr_at_100": 0.9524999999999999, | |
| "naucs_at_1_max": 0.5758636788048542, | |
| "naucs_at_1_std": -0.06209150326797414, | |
| "naucs_at_1_diff1": 0.9142156862745094, | |
| "naucs_at_3_max": 1.0, | |
| "naucs_at_3_std": 0.3384687208216551, | |
| "naucs_at_3_diff1": 1.0, | |
| "naucs_at_5_max": 1.0, | |
| "naucs_at_5_std": 0.12278244631185926, | |
| "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": null, | |
| "naucs_at_50_std": null, | |
| "naucs_at_50_diff1": null, | |
| "naucs_at_100_max": null, | |
| "naucs_at_100_std": null, | |
| "naucs_at_100_diff1": null | |
| }, | |
| "vidore/syntheticDocQA_healthcare_industry_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": 0.99, | |
| "mrr_at_3": 0.995, | |
| "mrr_at_5": 0.995, | |
| "mrr_at_10": 0.995, | |
| "mrr_at_20": 0.995, | |
| "mrr_at_50": 0.995, | |
| "mrr_at_100": 0.995, | |
| "naucs_at_1_max": 0.7222222222222248, | |
| "naucs_at_1_std": -0.14122315592903503, | |
| "naucs_at_1_diff1": 0.9346405228758133, | |
| "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": null, | |
| "naucs_at_50_std": null, | |
| "naucs_at_50_diff1": null, | |
| "naucs_at_100_max": null, | |
| "naucs_at_100_std": null, | |
| "naucs_at_100_diff1": null | |
| } | |
| } | |