tensorrt-cu13

What it is and what it does

tensorrt-cu13 is NVIDIA's Python interface to TensorRT, a specialized inference engine that takes trained neural network models and optimizes them for execution on NVIDIA GPUs. It compiles models from frameworks like ONNX and PyTorch into optimized TensorRT engines, reducing latency and memory footprint during inference. The package depends on tensorrt_cu13_libs (the compiled runtime) and tensorrt_cu13_bindings (the Python API layer), making it a wrapper around NVIDIA's native libraries rather than a pure-Python implementation.

The library is designed for production inference workloads where speed and efficiency matter—typical use cases include serving models in data centers, edge devices, or real-time applications. Version 11.2.1.2 represents a major release with breaking changes: weakly-typed networks, implicit quantization, and IPluginV2 plugins have been removed in favor of newer APIs. Installation requires CUDA 13 and a compatible GPU; it is not suitable for CPU-only environments.

Use it for:

• Compile ONNX or PyTorch models into optimized TensorRT engines for low-latency GPU inference in production
• Deploy large language models or vision models on NVIDIA data center GPUs with reduced memory and latency
• Benchmark and profile inference performance of neural networks on specific GPU hardware
• Integrate model optimization into CI/CD pipelines for automated inference engine generation
• Run inference on edge GPUs (Jetson devices) by leveraging TensorRT's memory-efficient compilation

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Provides Python bindings for NVIDIA TensorRT, a deep learning inference library that compiles and optimizes neural network models for deployment on NVIDIA GPUs.

Yes, if you need to optimize neural network inference on NVIDIA GPUs and have CUDA 13 and compatible hardware available. High install friction and proprietary licensing require upfront setup and legal review, but the package is actively maintained, widely used (top 15000 PyPI), and has no known vulnerabilities. Not suitable for CPU-only or non-NVIDIA GPU environments.

Install

tensorrt-cu13 on PyPI

pip install tensorrt-cu13

uv add tensorrt-cu13

poetry add tensorrt-cu13

Installing tensorrt-cu13

Before you install

High install friction due to compiled dependencies (tensorrt_cu13_libs and tensorrt_cu13_bindings) and CUDA 13 requirements. Package is actively maintained with recent releases, but installation complexity may require system-level CUDA setup.

License in practice

Licensed under Proprietary terms with unclear treatment. Users should verify licensing compliance with NVIDIA before deploying in production environments, particularly for commercial use.

Quickstart

pip install tensorrt-cu13

import tensorrt as trt

logger = trt.Logger(trt.Logger.WARNING)
builder = trt.Builder(logger)
network = builder.create_network(1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH))

Requires CUDA 13, NVIDIA GPU, and system-level CUDA toolkit installation. Python >= 3.10 required (despite requires_python claiming 3.8+).

Verify before relying

• Whether the proprietary license permits redistribution or modification of compiled models
• Specific CUDA version compatibility constraints beyond the cu13 suffix
• Whether Python 3.8 and 3.9 support claims in requires_python conflict with the description's removal of bindings for Python 3.9 and older

Package facts

Evidence: tensorrt_cu13-11.2.1.2.tar.gz

Keywords: nvidia, tensorrt, deeplearning, inference

Tags

More Artificial Intelligence packages

Further reading