How to adjust the model_list when changing the number of parities

[zjujy]

I tried to change k(the number of parities) from 2 to 4, and simply config the model as follows (copy the original model):
{
"epochs": 1,
"lr": 0.01,
"k": 4,
"batch_size": 1024,
"communication": {
"communication_protocol": "FedSGD",
"iteration_per_aggregation": 1
},
"dataset": {
"dataset_name": "mnist",
"num_classes": 10
},
"model_list": {
"0": {
"type": "MLP2",
"input_dim": 392,
"output_dim": 10,
"path": "random_1428_10"
},
"1": {
"type": "MLP2",
"input_dim": 392,
"output_dim": 10,
"path": "random_1428_10"
},
"2": {
"type": "MLP2",
"input_dim": 392,
"output_dim": 10,
"path": "random_1428_10"
},
"3": {
"type": "MLP2",
"input_dim": 392,
"output_dim": 10,
"path": "random_1428_10"
},
"apply_trainable_layer": 1,
"global_model": "ClassificationModelHostTrainableHead"
},
"defense": {
"name": "GaussianDP",
"parameters": {
"party": [
1
],
"dp_strength": 0.0001
}
},
"attack_list": {
"0": {
"name": "BatchLabelReconstruction",
"parameters": {
"party": [
0
],
"lr": 0.05,
"epochs": 10000
}
},
"1": {
"name": "DirectLabelScoring",
"parameters": {
"party": [
0
]
}
}
}
}
then I found the matrix shapes don't match

So I change the matrix shapes for all 4 models from 392*10 to 192 * 10, but it still does't work.

Could you please tell me how to modify the model list, thanks!