Server global evaluation total number

Below is a eval record of server. I wonder why test_total and val_total are decimals? From my understanding, this is global model evaluation. If the total number of my test set is 872, the number of test samples in each evaluation is 872 or 872/num of clients (in my case, it is 3)?

{'Role': 'Server #', 'Round': 1, 'Results_weighted_avg': {'test_loss': 205.44600348516343, 'test_total': 290.6666666666667, 'test_avg_loss': 0.7068027881307339, 'test_acc': 0.5091743119266054, 'val_loss': 154.90539731894472, 'val_total': 224.66666666666666, 'val_avg_loss': 0.6894755830396885, 'val_acc': 0.5534124629080118}, 'Results_avg': {'test_loss': 205.44401041666666, 'test_total': 290.6666666666667, 'test_avg_loss': 0.7067977845605226, 'test_acc': 0.5091875024686969, 'val_loss': 154.90218098958334, 'val_total': 224.66666666666666, 'val_avg_loss': 0.6894642857142856, 'val_acc': 0.5534457671957672}, 'Results_fairness': {'test_total': 290.6666666666667, 'val_total': 224.66666666666666, 'test_loss_std': 2.7915573098698245, 'test_loss_bottom_decile': 203.2431640625, 'test_loss_top_decile': 209.3828125, 'test_loss_min': 203.2431640625, 'test_loss_max': 209.3828125, 'test_loss_bottom10%': nan, 'test_loss_top10%': 209.3828125, 'test_loss_cos1': 0.9999076969612101, 'test_loss_entropy': 1.0985202582012823, 'test_avg_loss_std': 0.009149269452723627, 'test_avg_loss_bottom_decile': 0.6984301170532646, 'test_avg_loss_top_decile': 0.7195285652920962, 'test_avg_loss_min': 0.6984301170532646, 'test_avg_loss_max': 0.7195285652920962, 'test_avg_loss_bottom10%': nan, 'test_avg_loss_top10%': 0.7195285652920962, 'test_avg_loss_cos1': 0.999916228188608, 'test_avg_loss_entropy': 1.0985287222378004, 'test_acc_std': 0.02519823611017372, 'test_acc_bottom_decile': 0.4742268041237113, 'test_acc_top_decile': 0.5326460481099656, 'test_acc_min': 0.4742268041237113, 'test_acc_max': 0.5326460481099656, 'test_acc_bottom10%': nan, 'test_acc_top10%': 0.5326460481099656, 'test_acc_cos1': 0.998777755685617, 'test_acc_entropy': 1.097375117975531, 'val_loss_std': 2.4467666521872635, 'val_loss_bottom_decile': 152.734375, 'val_loss_top_decile': 158.32177734375, 'val_loss_min': 152.734375, 'val_loss_max': 158.32177734375, 'val_loss_bottom10%': nan, 'val_loss_top10%': 158.32177734375, 'val_loss_cos1': 0.9998752734853387, 'val_loss_entropy': 1.0984879472148361, 'val_avg_loss_std': 0.010041464909396417, 'val_avg_loss_bottom_decile': 0.6818498883928571, 'val_avg_loss_top_decile': 0.70365234375, 'val_avg_loss_min': 0.6818498883928571, 'val_avg_loss_max': 0.70365234375, 'val_avg_loss_bottom10%': nan, 'val_avg_loss_top10%': 0.70365234375, 'val_avg_loss_cos1': 0.9998939595599015, 'val_avg_loss_entropy': 1.0985065869345862, 'val_acc_std': 0.026944572693224266, 'val_acc_bottom_decile': 0.5155555555555555, 'val_acc_top_decile': 0.5758928571428571, 'val_acc_min': 0.5155555555555555, 'val_acc_max': 0.5758928571428571, 'val_acc_bottom10%': nan, 'val_acc_top10%': 0.5758928571428571, 'val_acc_cos1': 0.9988169822369142, 'val_acc_entropy': 1.0974135282771185}}

The issue provides a server evaluation record but names no source files, tests, or entry points. Locate the server evaluation aggregation that produces test_total and val_total, then verify whether those values are averaged across clients or represent global sample counts; done means clarifying or correcting that behavior and validating the reported totals.

Written by the indexing model from the issue text.