NVIDIA NVJITLink

13.2.51 · active · verified Thu Apr 09

NVIDIA NVJITLink is a Python package that distributes the native NVIDIA NVJITLink shared library (libnvjitlink). This library provides JIT (Just-In-Time) compilation services for NVIDIA GPU code, acting as a low-level component within the NVIDIA ecosystem. It's primarily used as a backend dependency by higher-level Python libraries and frameworks (e.g., Numba, CuPy, PyTorch with NVFuser) that need to compile GPU kernels at runtime. Its version is often tightly coupled with the NVIDIA CUDA Toolkit version. The current version is 13.2.51, and it's updated in conjunction with CUDA Toolkit releases.

Warnings

gotcha NVJITLink is a low-level backend library. It is not intended for direct use by application developers to write GPU kernels. Its primary function is to provide JIT compilation services to other higher-level frameworks like Numba, CuPy, or PyTorch.
Fix: Do not attempt to write GPU code directly using `nvidia-nvjitlink`. Instead, use libraries built on top of it, such as Numba for JIT compilation, or PyTorch/CuPy for array-based GPU computation.
gotcha The version of `nvidia-nvjitlink` is often tied to a specific NVIDIA CUDA Toolkit version. Mismatches between the installed `nvidia-nvjitlink` package version and the system's CUDA driver or toolkit version can lead to runtime errors or unexpected behavior in dependent libraries.
Fix: Ensure that the `nvidia-nvjitlink` version is compatible with your CUDA driver and toolkit. When using dependent libraries (e.g., Numba, CuPy), always check their recommended installation procedures and dependencies, as they often handle `nvidia-nvjitlink` compatibility implicitly.
gotcha Many Python GPU-accelerated libraries (e.g., `numba`, `cupy`) list `nvidia-nvjitlink` as a dependency and will install it automatically. Directly installing `nvidia-nvjitlink` might not be necessary and could potentially lead to version conflicts if the dependent library requires a specific version.
Fix: Prioritize installing the higher-level GPU library you intend to use. Allow it to manage `nvidia-nvjitlink` as its dependency. Only install `nvidia-nvjitlink` directly if explicitly instructed or if you are developing a low-level library that specifically requires direct access to `libnvjitlink`.

Install

pip install nvidia-nvjitlink Install latest version

Imports

__version__
```
import nvidia_nvjitlink; print(nvidia_nvjitlink.__version__)
```
nvidia-nvjitlink is primarily a backend library; direct Python API usage by end-users for functionality is uncommon. The Python package mainly serves to distribute the native `libnvjitlink` shared library. Checking `__version__` is the most common direct interaction to confirm installation.

Quickstart

This quickstart demonstrates how to import the `nvidia_nvjitlink` package and access its `__version__` attribute, which is the most common direct interaction for end-users. It verifies the package's presence and version, confirming it's available for other GPU-accelerated libraries that depend on it.

import nvidia_nvjitlink

print(f"NVIDIA NVJITLink package version: {nvidia_nvjitlink.__version__}")
# NVJITLink is a backend library; direct high-level API usage is not typical.
# This quickstart confirms the package is installed and its version.

view raw JSON →