Implementation:LaurentMazare Tch rs VarStore New
| Knowledge Sources | |
|---|---|
| Domains | Deep_Learning, Parameter_Management |
| Last Updated | 2026-02-08 14:00 GMT |
Overview
Concrete tool for creating a device-bound parameter store provided by the tch nn module.
Description
nn::VarStore::new creates a new variable store on the specified device. The VarStore holds an Arc<Mutex<Variables>> containing an empty named_variables HashMap and an empty trainable_variables Vec. The default kind is Float (f32). All subsequent layer constructors that receive a path from this VarStore will register their parameters into it.
Usage
Call this as the first step when building any neural network model. Pass Device::Cpu for CPU training or Device::cuda_if_available() for GPU training. The VarStore must be created before any layers.
Code Reference
Source Location
- Repository: tch-rs
- File: src/nn/var_store.rs
- Lines: 58-62
Signature
pub fn new(device: Device) -> VarStore
Import
use tch::nn;
use tch::Device;
I/O Contract
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
| device | Device | Yes | Target compute device (Device::Cpu, Device::Cuda(0), or Device::cuda_if_available()) |
Outputs
| Name | Type | Description |
|---|---|---|
| VarStore | nn::VarStore | Empty variable store with no parameters, default Kind::Float |
Usage Examples
CPU VarStore
use tch::{nn, Device};
let vs = nn::VarStore::new(Device::Cpu);
let root = vs.root(); // Get root path for layer construction
CUDA VarStore
use tch::{nn, Device};
// Automatically use CUDA if available, otherwise CPU
let vs = nn::VarStore::new(Device::cuda_if_available());
println!("Training on: {:?}", vs.device());
Related Pages
Implements Principle
Requires Environment
Uses Heuristic
Page Connections
Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment