How to Call PyTorch from Rust via FFI

Unsafe Ffi
2026-04-17T12:05:42+00:00
You cannot call PyTorch directly from Rust via FFI because PyTorch is a C++ library with no stable C API; you must use a Rust binding crate like `pyo3` to call Python PyTorch or `torch-sys` to call the C++ core. Use `pyo3` to run Python code from Rust, which is the standard approach for accessing Py

How to Call PyTorch from Rust via FFI

You cannot call PyTorch directly from Rust via FFI because PyTorch is a C++ library with no stable C API; you must use a Rust binding crate like pyo3 to call Python PyTorch or torch-sys to call the C++ core. Use pyo3 to run Python code from Rust, which is the standard approach for accessing PyTorch models.

use pyo3::prelude::*;

#[pyfunction]
fn run_pytorch_model(py: Python) -> PyResult<f64> {
    let torch = py.import("torch")?;
    let model = torch.getattr("load")?.call1(("model.pt",))?;
    // Call model inference here
    Ok(0.95)
}

#[pymodule]
fn my_module(_py: Python, m: &PyModule) -> PyResult<()> {
    m.add_function(wrap_pyfunction!(run_pytorch_model, m)?)?;
    Ok(())
}

Add pyo3 = { version = "0.20", features = ["extension-module"] } to your Cargo.toml.