Simple Sequence Model Helpers (API)

Small, reusable constructors for “sequence core + time-distributed linear head” models used by the rnn and lstm runnable examples.

These helpers live in the API layer so examples can stay focused on: corpus loading, sample construction, and training loops.

structure NN.API.nn.models.SeqRnnHeadConfig : Type

Configuration for an RNN/LSTM sequence model with a time-distributed linear head.

Shapes follow the convention used by the runnable examples:

• input: (seqLen × inputSize)
• output logits/targets: (seqLen × inputSize)

• seqLen : ℕ
• inputSize : ℕ
• hiddenSize : ℕ

def NN.API.nn.models.instReprSeqRnnHeadConfig.repr : SeqRnnHeadConfig → ℕ → Std.Format

@[implicit_reducible]

instance NN.API.nn.models.instReprSeqRnnHeadConfig : Repr SeqRnnHeadConfig

@[reducible, inline]

abbrev NN.API.nn.models.seqRnnHeadInShape (cfg : SeqRnnHeadConfig) : Shape

Input shape (seqLen × inputSize) for SeqRnnHeadConfig.

@[reducible, inline]

abbrev NN.API.nn.models.seqRnnHeadOutShape (cfg : SeqRnnHeadConfig) : Shape

Output shape (seqLen × inputSize) for SeqRnnHeadConfig.

def NN.API.nn.models.rnnWithLinearHead (cfg : SeqRnnHeadConfig) : M (Sequential (seqRnnHeadInShape cfg) (seqRnnHeadOutShape cfg))

Vanilla RNN core plus time-distributed linear head:

rnn(seqLen,inputSize,hiddenSize) → linear(hiddenSize → inputSize).

def NN.API.nn.models.lstmWithLinearHead (cfg : SeqRnnHeadConfig) : M (Sequential (seqRnnHeadInShape cfg) (seqRnnHeadOutShape cfg))

LSTM core plus time-distributed linear head:

lstm(seqLen,inputSize,hiddenSize) → linear(hiddenSize → inputSize).