Doubts about ParallelTable and ParallelCriterion

[clare-cn]

I am a beginner using BigDL to create a ParallelTable model for training, but I encountered errors in converting between Tensor and Table types:
Caused by: java. lang. ClassCastException: com. intel. analytics. bigdl. dllib. utilities Table cannot be cast to com.intel.analytics.bigdl.dllib.tensor.Tensor.
I have used Graph (Input [Float(), Sigmoid [Float())) models before, with a loss function of MSERiterion [Float()) and training data input of DataSet [Sample [Float] (Tensor (features, Array (features. length), Tensor (Array (label), Array (1)), can work normally. But now, I am confused about this. Is there a problem with my training data type (but Sample only accepts Tensor types)? Or is there a problem with my model construction? I don't know how to solve it now, I hope to receive some help
My partial code is as follows:

val modelTask1 = Sequential()
.add(SequentialFloat
.add(Linear(200, 100))
.add(ReLUFloat))
.add(Linear(100, 1))
val modelTask2 = Sequential()
.add(SequentialFloat
.add(Linear(200, 100))
.add(ReLUFloat))
.add(Linear(100, 1))

val model = ParallelTable()
.add(modelTask1)
.add(modelTask2)
val trainData = DataSet.rdd(trainDf.rdd.map(row => {
val features = row.getAsDenseVector.toArray.map(_.toFloat)
val ctr = row.getAsDouble.toFloat
val cvr = row.getAsDouble.toFloat
Sample[Float](Tensor(features, Array(features.length)), Tensor(Array(ctr, cvr), Array(2)))
})).transform(SampleToMiniBatchFloat)

val criterion = ParallelCriterionFloat
criterion.add(MSECriterionFloat, 1.0)
criterion.add(MSECriterionFloat, 1.0)
val optimizer = Optimizer(
model = model,
dataset = trainData,
criterion = criterion
).setOptimMethod(new AdamFloat)
.setEndWhen(Trigger.maxEpoch(maxEpoch.toInt))
val trainedModel = optimizer.optimize()