load partitions for matrix fclayer_1012_1_weight failed, error log is LOAD model partitions failed.-4998

使用com.tencent.angel.spark.ml.core.GraphModel，init之后用com.tencent.angel.spark.ml.core.OfflineLearner train，然后调用GraphModel的save接口把模型保存下来，再用GraphModel load的时候就报这个错误了。
save下来的json内容如下：
``` json
{
"data": {
"format": "libsvm",
"indexrange": 148,
"numfield": 148,
"validateratio": 0.0,
"sampleratio": 1.0
},
"model": {
"modeltype": "T_DOUBLE_DENSE",
"modelsize": -1,
"blockSize": 1000000,
"numClass": 2
},
"train": {
"epoch": 10,
"numupdateperepoch": 10,
"batchsize": 128,
"lr": 0.1,
"decayclass": "StandardDecay",
"decayalpha": 0.001,
"decaybeta": 0.001
},
"layers": [
{
"name": "wide_1012",
"type": "SimpleInputLayer",
"outputdim": 1,
"transfunc": {
"type": "Identity"
},
"optimizer": {
"type": "Adam",
"beta": 0.9,
"gamma": 0.99
}
},
{
"name": "embedding_1012",
"type": "Embedding",
"outputdim": 1184,
"numfactors": 8,
"optimizer": {
"type": "Adam",
"beta": 0.9,
"gamma": 0.99
}
},
{
"name": "fclayer_1012",
"type": "FCLayer",
"outputdims": [
100,
50,
1
],
"inputlayer": "embedding_1012",
"transfuncs": [
{
"type": "Relu"
},
{
"type": "Relu"
},
{
"type": "Identity"
}
],
"optimizer": {
"type": "Adam",
"beta": 0.9,
"gamma": 0.99
}
},
{
"name": "sumPooling_1012",
"type": "SumPooling",
"outputdim": 1,
"inputlayers": [
"wide_1012",
"fclayer_1012"
]
},
{
"name": "simplelosslayer_1012",
"type": "SimpleLossLayer",
"outputdim": -1,
"inputlayer": "sumPooling_1012",
"lossfunc": {
"type": "LogLoss"
}
}
]
}
```
使用的angel版本是3.2.0 af8ebe8，然而我自己的代码在未作任何修改的情况下，整个流程使用ange版本3.1.0是没问题的，当时的commit id是 ee20d05