---

name: sequence-and-rnn description: "Use Sonnet RNN cores, sequence unroll helpers, trainable recurrent state, and convolutional LSTMs." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Sequence and RNN

Use this sub-skill for snt.RNNCore, VanillaRNN, LSTM, GRU, DeepRNN, skip/residual helpers, static_unroll, dynamic_unroll, TrainableState, UnrolledLSTM, recurrent dropout, and ConvLSTM modules.

Start here

• references/api-reference.md: signatures, state structures, and shape conventions.
• references/workflows.md: LSTM/GRU unrolls, DeepRNN, trainable state, recurrent dropout, and ConvLSTM recipes.
• references/troubleshooting.md: time-major, state, skip/residual, ConvLSTM, and backend-specialization errors.
• scripts/rnn_unroll_smoke.py: safe LSTM/GRU unroll shape smoke.

Boundaries

• Generic snt.Module authoring: ../module-authoring/SKILL.md.
• Feed-forward layers/nets: ../layers-and-nets/SKILL.md.
• Optimizer loops: ../training-and-optimization/SKILL.md.
• XLA/distribution/backend policy: ../serialization-and-distribution/SKILL.md.