Training #2

Tang-RNN

SetDirectory[NotebookDirectory[]];
encoder = NetEncoder[{"Characters", chars = Append[Import["chars.WXF"], _]}]
trainingData = encoder /@ Import["ipt.WXF"]; // TT
predictNet = NetChain[{
	UnitVectorLayer[Length@chars],
	GatedRecurrentLayer[128],
	GatedRecurrentLayer[128],
	NetMapOperator[LinearLayer[Length@chars]],
	SoftmaxLayer[]
}];
teacherForcingNet = NetGraph[<|
	"predict" -> predictNet,
	"rest" -> SequenceRestLayer[],
	"most" -> SequenceMostLayer[],
	"loss" -> CrossEntropyLossLayer["Index"]
|>, {
	NetPort["Input"] -> "most" -> "predict" -> NetPort["loss", "Input"],
	NetPort["Input"] -> "rest" -> NetPort["loss", "Target"]
},
	"Input" -> {Length@First[trainingData], "Integer"}
] // NetInitialize
result = NetTrain[teacherForcingNet,
	<|"Input" -> trainingData|>, All,
	MaxTrainingRounds -> 100, TimeGoal -> 3600,
	TrainingProgressCheckpointing -> {
		"Directory", "CheckPoints", "Interval" -> Quantity[10000, "Batches"]
	},
	BatchSize -> 64, TargetDevice -> "GPU",
	ValidationSet -> Scaled[0.01]
]
Export["result.WXF", result, PerformanceGoal -> "Size"]

Training #1

REPORT = <|
	"Quality" -> {
		<|"FinalRoundErrorRate" -> 0.42033032319391633|>,
		<|"FinalRoundLoss" -> 1.355493026017108|>,
		<|"FinalValidationErrorRate" -> 0.43569057901471525|>,
		<|"FinalValidationLoss" -> 1.4399253633338063|>,
		<|"LowestValidationErrorRate" -> 0.43569057901471525|>,
		<|"LowestValidationLoss" -> 1.4399253633338063|>,
		<|"LowestValidationRound" -> 26|>
	},
	"Quantity" -> {
		<|"TotalTrainingTime" -> 3600.0066981966856|>,
		<|"TotalBatches" -> 351813|>,
		<|"TotalInputs" -> 22516032|>,
		<|"TotalRounds" -> 26|>,
		<|"MeanBatchesPerSecond" -> 97.72565150399026|>,
		<|"MeanInputsPerSecond" -> 6254.441696255377|>
	},
	"Method" -> {
		<|"OptimizationMethod" -> "ADAM"|>,
		<|"InitialLearningRate" -> 0.001|>,
		<|"FinalLearningRate" -> 0.001|>,
		<|"BatchSize" -> 64|>,
		<|"VersionNumber" -> "11.3.5"|>
	}
|>;

First Blood!

v1.0.0

模型::Tableaux remake