JYPPX.TensorRT.CSharp.API 4.0.0-preview.1

<p align="center"> <img src="https://socialify.git.ci/guojin-yan/TensorRT-CSharp-API/image?description=1&descriptionEditable=TensorRT%20and%20CUDA%20bindings%20for%20C%23%20and%20.NET&forks=1&issues=1&name=1&owner=1&pattern=Circuit%20Board&pulls=1&stargazers=1&theme=Light" alt="TensorRtSharp4.0" width="100%" /> </p>

<h1 align="center">TensorRtSharp4.0</h1>

<p align="center"> TensorRT and CUDA bindings for C# and .NET, with managed APIs, project-owned bridge packages, runnable vision samples, and a TensorRtExec desktop workflow. </p>

<p align="center"> <a href="LICENSE"><img src="https://img.shields.io/badge/License-Apache%202.0-blue.svg" alt="Apache-2.0 License" /></a> <a href="https://www.nuget.org/packages/JYPPX.TensorRT.CSharp.API/"><img src="https://img.shields.io/nuget/vpre/JYPPX.TensorRT.CSharp.API.svg" alt="NuGet prerelease version" /></a> <a href="https://www.nuget.org/packages/JYPPX.TensorRT.CSharp.API/"><img src="https://img.shields.io/nuget/dt/JYPPX.TensorRT.CSharp.API.svg" alt="NuGet downloads" /></a> <a href="https://github.com/users/guojin-yan/packages/nuget/package/jyppx.tensorrt.csharp.api"><img src="https://img.shields.io/badge/GitHub%20Packages-package%20feed-24292f" alt="GitHub Packages feed" /></a> <a href="https://github.com/guojin-yan/TensorRT-CSharp-API/releases"><img src="https://img.shields.io/github/v/release/guojin-yan/TensorRT-CSharp-API?include_prereleases&label=Release" alt="GitHub Release" /></a> <a href="https://dotnet.microsoft.com/"><img src="https://img.shields.io/badge/.NET-Framework%204.6--4.8.1%20%7C%20Core%203.1%20%7C%205--10-512BD4" alt="Supported .NET versions" /></a> <a href="https://developer.nvidia.com/tensorrt"><img src="https://img.shields.io/badge/TensorRT-user%20installed-76B900" alt="TensorRT is user installed" /></a> </p>

<p align="center"> <a href="https://github.com/guojin-yan/TensorRT-CSharp-API/actions/workflows/ci-validation.yml"><img src="https://github.com/guojin-yan/TensorRT-CSharp-API/actions/workflows/ci-validation.yml/badge.svg?branch=TensorRtSharp4.0" alt="Managed CI" /></a> <a href="https://github.com/guojin-yan/TensorRT-CSharp-API/actions/workflows/release-quality-gate.yml"><img src="https://github.com/guojin-yan/TensorRT-CSharp-API/actions/workflows/release-quality-gate.yml/badge.svg?branch=TensorRtSharp4.0" alt="Release quality gate" /></a> </p>

<p align="center"><strong>English</strong> | <a href="README.zh-CN.md">简体中文</a></p>

TensorRtSharp4.0

TensorRtSharp4.0 provides a .NET API for TensorRT inference, CUDA runtime compilation, memory, streams, callbacks, and the TensorRtExec desktop workflow. The first public candidate is planned as <code>4.0.0-preview.1</code>; it is a source and package candidate, not a claim that every TensorRT feature has completed runtime validation.

Introduction

The managed API keeps the public namespace roots stable:

• <code>JYPPX.TensorRtSharp</code> is the main TensorRT surface and the default home for shared types.
• <code>JYPPX.CudaSharp</code> is the CUDA surface.
• Native bridge loading is explicit through <code>jyppxtrtbridge</code> and the versioned bridge package.

CUDA, cuDNN, TensorRT, and NVRTC are user-installed prerequisites. NVIDIA runtime redistribution is retired: this repository publishes managed source/packages and project-owned bridge-only packages, never CUDA/cuDNN/TensorRT vendor archives.

Release Highlights

• C# bindings and source are grouped by module under <code>src</code>.
• TensorRT engine building, execution contexts, bindings, dynamic-shape profiles, allocators, logging, profiling, progress monitoring, streams, events, CUDA graphs, and CUDA RTC are covered by the managed API.
• Runtime package roles are explicit: the .Bridge packages with split_package_roles=bridge contain only the project bridge for a fixed CUDA/TensorRT line.
• NuGet branding is fixed: <code>nuget/logo.jpg</code> is the package logo and the English root README is embedded as the package README.
• The release license is Apache-2.0.

Get Started In 30 Seconds

Create a console project, reference the managed package, and install the bridge package that matches the CUDA/TensorRT installation on the target machine:

dotnet new console -n TrtQuickstart
cd TrtQuickstart
dotnet add package JYPPX.TensorRT.CSharp.API --version 4.0.0-preview.1
dotnet add package JYPPX.TensorRT.CSharp.API.Bridge.win-x64-trt10.11-cuda12.9-cudnn9.22 --version 4.0.0-preview.1

Then create a runtime, load an engine, bind input/output tensors, execute, and read the result. The bridge package is not a replacement for the user-installed NVIDIA runtime. See the inference bindings tutorial and Windows installation guide.

Package Layout

samples/YoloVision and samples/Classification are runnable demonstrations. Their project files and local package-consumer harnesses remain in the source tree for development checks, but their sample package IDs are deliberately excluded from all public package feeds and Release assets.

Public Packages And Release Assets

The first public candidate is <code>4.0.0-preview.1</code>. Package README content is the English root README, package branding uses <code>nuget/logo.jpg</code>, and the core managed package uses the Apache-2.0 SPDX license expression.

Package	Version	NuGet.org	GitHub Packages	Purpose
<code>JYPPX.TensorRT.CSharp.API</code>		Gallery	Package feed	Core managed TensorRT/CUDA API

Bridge package matrix

Every public bridge package is listed below. CUDA, cuDNN, and TensorRT are prerequisites installed by the consumer; the .Bridge package contains only jyppxtrtbridge.

Runtime packages do not bundle NVIDIA libraries. For local source builds, use the scripts in <code>eng</code> only through the documented entry points; most exporter and owner-proof scripts are internal engineering tools.

Native Dependencies

Install the matching NVIDIA stack before running a bridge package:

Use <code>TENSORRT_PATH</code>, <code>JYPPX_TENSORRT_ROOT</code>, and the platform loader path appropriate for your machine. The repository does not upload or package these vendor runtimes.

Models And ONNX Conversion

Demo models are staged outside Git in the sibling <code>models</code> directory and are not included in source archives or packages. Each article records the official acquisition URL, pinned revision, license, conversion command, input/output contract, and SHA256.

See the demo model inventory, acquisition and conversion guide, and <code>eng/Sync-DemoOnnxModels.ps1</code>. Model files stay in the external model store until ModelZoo is available.

Documentation

• English documentation
• Chinese article catalog and publication criteria
• Project overview
• Source organization
• Model acquisition and ONNX conversion
• Inference bindings
• TensorRtExec GUI
• Release candidate gate
• Release proof sample article
• Release proof owner dashboard
• TensorRtExec report boundary
• ONNX to engine report boundary
• YOLOVision asset evidence guide
• Callback/allocator safety gates
• API readiness audit
• YOLOVision model matrix
• TensorRtExec feature matrix
• ONNX-to-engine parity matrix
• Article roadmap 30-plus

Build From Source

dotnet restore TensorRtSharp.sln
dotnet build TensorRtSharp.sln -c Release
dotnet test tests/JYPPX.ProjectQuality.Tests/JYPPX.ProjectQuality.Tests.csproj -c Release --no-restore

Pack locally with <code>JYPPXPackageVersion=4.0.0-preview.1</code>. Inspect the generated nupkg before any upload; it must contain the package README and <code>logo.jpg</code> and must not contain CUDA, cuDNN, or TensorRT vendor binaries.

Release And Action Policy

Workflows are manual-only to conserve Actions quota. The <code>grape-yan</code> repository is validation-only: it may receive one candidate Action after local checks pass and never publishes. The formal <code>guojin-yan</code> workflow is dispatched once, with explicit Owner approval, for the fixed first version <code>4.0.0-preview.1</code>.

Until that run completes, release artifacts remain <code>blocked</code>. The state is <code>owner-action-required</code>. <code>clean-consumer-proof-execution-bundle</code> and <code>clean-consumer-external-proof-closure-pack</code> are <code>non-proof</code> Owner actions; they do not run runtime smoke, are not runtime proof, and are not post-publish proof. Build-only output, local feeds, ProjectReference, dry-runs, templates, and dashboards are rejected by <code>FailOnNotProof</code> and cannot be promoted to publication or issue close.

NuGet publication requires push permission for JYPPX.TensorRT.CSharp.API only. JYPPX.TensorRT.CSharp.API.YoloVision and JYPPX.TensorRT.CSharp.API.Classification are sample-only IDs and must never be uploaded. A nuget.org 403 is an authorization failure, not a retryable build failure.

Repository Layout

• <code>src</code>: managed interfaces grouped by CUDA, TensorRT, runtime, memory, and shared modules.
• <code>native</code>: project bridge source and ABI exports.
• <code>samples</code>: runnable C# demonstrations and model metadata.
• <code>applications/TensorRtExec</code>: desktop engine builder and runner.
• <code>pack</code>: managed and bridge-only package definitions.
• <code>docs</code>: DocFX site and technical articles.
• <code>eng</code>: build, acquisition, validation, and release engineering scripts.

License

Licensed under Apache-2.0. See LICENSE.

Support

Please include the package version, CUDA/cuDNN/TensorRT versions, GPU model, operating system, and the failing command when opening an issue. Do not upload proprietary model weights or NVIDIA runtime archives. *** End Patch