TorchLean API

Docs HomeInstallationGuideExamplesGraphsTools

NN.API.Models.Mlp

Multilayer Perceptrons #

Compact, reusable MLP-shaped model constructors for TorchLean examples.

These live in the API layer so runnable examples can stay focused on: data loading, training loops, and CLI flags, rather than repeating the same linear → activation → linear boilerplate.

Scope note: these are building blocks, not pretrained checkpoints.

source
structure TorchLean.nn.models.MlpConfig :
Type

Configuration for a single-hidden-layer MLP over batched row vectors.

Instances For
    source
    def TorchLean.nn.models.instReprMlpConfig.repr :
    MlpConfig → ℕ → Std.Format
    Instances For
      source
      @[implicit_reducible]
      instance TorchLean.nn.models.instReprMlpConfig :
      Repr MlpConfig
      source
      @[reducible, inline]
      abbrev TorchLean.nn.models.mlpInShape (cfg : MlpConfig) :
      Spec.Shape

      Input shape (batch × inDim) for an MlpConfig.

      Instances For
        source
        @[reducible, inline]
        abbrev TorchLean.nn.models.mlpOutShape (cfg : MlpConfig) :
        Spec.Shape

        Output shape (batch × outDim) for an MlpConfig.

        Instances For
          source
          def TorchLean.nn.models.mlpRelu (cfg : MlpConfig) :
          M (Sequential (mlpInShape cfg) (mlpOutShape cfg))

          Build a single-hidden-layer MLP with relu activation:

          linear(inDim → hidDim) → relu → linear(hidDim → outDim).

          Instances For