Implementation:LaurentMazare Tch rs Nn Seq

Knowledge Sources	tch-rs
Domains	Deep_Learning, Model_Architecture
Last Updated	2026-02-08 14:00 GMT

Overview

Concrete tool for building sequential layer chains provided by the tch nn module.

Description

nn::seq creates an empty Sequential container. Layers are added via the builder pattern using .add() for Module layers and .add_fn() for closure-based transformations. The Sequential container implements the Module trait, enabling it to be used as a single layer in larger architectures. A SequentialT variant is available for layers requiring the train flag (e.g., dropout, batch norm).

Usage

Use this when constructing feedforward architectures. Call nn::seq() then chain .add() calls to compose your network.

Code Reference

Source Location

• Repository: tch-rs
• File: src/nn/sequential.rs
• Lines: 12-14

Signature

pub fn seq() -> Sequential

Import

use tch::nn;

I/O Contract

Inputs

Name	Type	Required	Description
(none)	—	—	No arguments; layers added via builder pattern

Outputs

Name	Type	Description
Sequential	nn::Sequential	Empty layer container implementing Module trait

Usage Examples

Two-Layer Feedforward Network

use tch::{nn, nn::Module};

let vs = nn::VarStore::new(tch::Device::Cpu);
let net = nn::seq()
    .add(nn::linear(vs.root() / "layer1", 784, 128, Default::default()))
    .add_fn(|xs| xs.relu())
    .add(nn::linear(vs.root() / "layer2", 128, 10, Default::default()));

// Forward pass
let input = tch::Tensor::randn([1, 784], tch::kind::FLOAT_CPU);
let output = net.forward(&input);  // shape: [1, 10]

Related Pages

Implements Principle

• Principle:LaurentMazare_Tch_rs_Sequential_Model_Definition