TorchLean API

Docs HomeGuideExamplesGraphs

NN.API.Public.Facade.Trainer.Summary

TorchLean Trainer Summaries #

Small public report types returned by the high-level trainer facade.

source
structure TorchLean.Trainer.TrainSummary :
Type

Backend-independent before/after training summary.

Losses are rendered as strings because the runtime scalar type is chosen inside the training run.

  • metric : String

    Metric name, usually loss.

  • steps : ℕ

    Number of optimizer steps requested by the configuration.

  • before : String

    Metric before training.

  • after : String

    Metric after training.

Instances For
    source
    def TorchLean.Trainer.TrainSummary.summary (report : TrainSummary) :
    String

    One-line summary suitable for quickstarts and scripts.

    Instances For
      source
      def TorchLean.Trainer.TrainSummary.printSummary (report : TrainSummary) :
      IO Unit

      Print the one-line before/after summary.

      Instances For
        source
        @[implicit_reducible]
        instance TorchLean.Trainer.TrainSummary.instToString :
        ToString TrainSummary
        source
        def TorchLean.Trainer.TrainSummary.parseFloat? (s : String) :
        Option Float

        Parse a ToString-rendered scalar as a JSON number when possible.

        Instances For
          source
          def TorchLean.Trainer.TrainSummary.toTrainLog? (title : String) (notes : Array String) (report : TrainSummary) :
          Option Training.TrainLog

          Convert a before/after summary into the standard two-point TrainLog when values are finite.

          Instances For
            source
            def TorchLean.Trainer.TrainSummary.requireFloatLosses (context : String) (report : TrainSummary) :
            IO (Float × Float)

            Read the before/after metrics back as ordinary Floats.

            Most scripts call trained.printSummary. Examples that write JSON logs can use this operation when they need the same metrics as Floats.

            Instances For
              source
              def TorchLean.Trainer.TrainSummary.requireAndPrintFloatLosses (context : String) (report : TrainSummary) (steps? : Option ℕ := none) (lr? : Option Float := none) :
              IO (Float × Float)

              Print before/after losses and return them for artifact writers.

              Instances For
                source
                def TorchLean.Trainer.TrainSummary.writeLog (dest : Training.LogDestination) (title : String) (notes : Array String) (report : TrainSummary) :
                IO Unit

                Write this summary to a log destination when logging is enabled.

                Instances For