Implementation:LaurentMazare Tch rs CModule Load

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

Overview

Concrete tool for loading TorchScript .pt models for inference in Rust provided by the tch wrappers module.

Description

CModule::load deserializes a TorchScript .pt file into a CModule wrapper around the C++ torch::jit::Module. The variant CModule::load_on_device additionally specifies the target device, allowing GPU-exported models to be loaded on CPU. The CModule implements the Module trait and provides forward_ts (tensor slice input) and forward_is (IValue input) methods.

Usage

Use for inference-only scenarios where you do not need to train the model in Rust. For training, use TrainableCModule instead.

Code Reference

Source Location

• Repository: tch-rs
• File: src/wrappers/jit.rs
• Lines: 435-439 (load), 445-452 (load_on_device)

Signature

impl CModule {
    pub fn load<T: AsRef<std::path::Path>>(path: T) -> Result<CModule, TchError>

    pub fn load_on_device<T: AsRef<std::path::Path>>(
        path: T,
        device: Device,
    ) -> Result<CModule, TchError>
}

Import

use tch::CModule;

I/O Contract

Inputs

Name	Type	Required	Description
path	T: AsRef<Path>	Yes	Path to TorchScript .pt file
device	Device	No	Target device (only for load_on_device)

Outputs

Name	Type	Description
Result<CModule>	CModule	Loaded JIT module implementing Module trait

Usage Examples

use tch::{CModule, Device, nn::Module, vision::imagenet};

// Load on default device
let model = CModule::load("resnet18.pt")?;

// Load GPU model on CPU
let model = CModule::load_on_device("resnet18.pt", Device::Cpu)?;

// Inference
let image = imagenet::load_image_and_resize224("photo.jpg")?;
let output = image.unsqueeze(0).apply(&model);

Related Pages

Implements Principle

• Principle:LaurentMazare_Tch_rs_TorchScript_Loading