Microsoft.ML.OnnxRuntimeGenAI.Cuda 0.1.0

The ID prefix of this package has been reserved for one of the owners of this package by Prefix Reserved
There is a newer version of this package available.
See the version list below for details.
dotnet add package Microsoft.ML.OnnxRuntimeGenAI.Cuda --version 0.1.0
NuGet\Install-Package Microsoft.ML.OnnxRuntimeGenAI.Cuda -Version 0.1.0
This command is intended to be used within the Package Manager Console in Visual Studio, as it uses the NuGet module's version of Install-Package.
<PackageReference Include="Microsoft.ML.OnnxRuntimeGenAI.Cuda" Version="0.1.0" />
For projects that support PackageReference, copy this XML node into the project file to reference the package.
paket add Microsoft.ML.OnnxRuntimeGenAI.Cuda --version 0.1.0
#r "nuget: Microsoft.ML.OnnxRuntimeGenAI.Cuda, 0.1.0"
#r directive can be used in F# Interactive and Polyglot Notebooks. Copy this into the interactive tool or source code of the script to reference the package.
// Install Microsoft.ML.OnnxRuntimeGenAI.Cuda as a Cake Addin
#addin nuget:?package=Microsoft.ML.OnnxRuntimeGenAI.Cuda&version=0.1.0

// Install Microsoft.ML.OnnxRuntimeGenAI.Cuda as a Cake Tool
#tool nuget:?package=Microsoft.ML.OnnxRuntimeGenAI.Cuda&version=0.1.0

ONNX Runtime Generative AI

Run generative AI models with ONNX Runtime.

This library provides the generative AI loop for ONNX models, including inference with ONNX Runtime, logits processing, search and sampling, and KV cache management.

Users can call a high level generate() method, or run each iteration of the model in a loop.

  • Support greedy/beam search and TopP, TopK sampling to generate token sequences
  • Built in logits processing like repetition penalties
  • Easy custom scoring


  • Supported model architectures:
    • Gemma
    • LLaMA
    • Mistral
    • Phi-2
  • Supported targets:
    • CPU
    • GPU (CUDA)
  • Supported sampling features
    • Beam search
    • Greedy search
    • Top P/Top K
  • APIs
    • Python
    • C/C++

Coming very soon

  • Support for the Whisper model architectures
  • C# API
  • Support for DirectML


  • Automatic model download and cache
  • More model architectures

Sample code for phi-2 in Python

Install onnxruntime-genai.

(Temporary) Build and install from source according to the instructions below.

import onnxruntime_genai as og

model = og.Model(f'models/microsoft/phi-2', device_type)

tokenizer = og.Tokenizer(model)

prompt = '''def print_prime(n):
    Print all primes between 1 and n

tokens = tokenizer.encode(prompt)

params = og.SearchParams(model)
params.input_ids = tokens

output_tokens = model.generate(params)

text = tokenizer.decode(output_tokens)


Build from source

This step requires cmake to be installed.

  1. Clone this repo

    git clone
    cd onnxruntime-genai
  2. Install ONNX Runtime

    By default, the onnxruntime-genai build expects to find the ONNX Runtime include and binaries in a folder called ort in the root directory of onnxruntime-genai. You can put the ONNX Runtime files in a different location and specify this location to the onnxruntime-genai build. These instructions use ORT_HOME as the location.

    • Install from release

      These instructions are for the Linux GPU build of ONNX Runtime. Replace the location with the operating system and target of choice.

      cd $ORT_HOME
      tar xvzf onnxruntime-linux-x64-gpu-1.17.1.tgz 
      mv onnxruntime-linux-x64-gpu-1.17.1/include .
      mv onnxruntime-linux-x64-gpu-1.17.1/lib .
    • Or build from source

      git clone
      cd onnxruntime

      Create include and lib folders in the ORT_HOME directory

      mkdir $ORT_HOME/include
      mkdir $ORT_HOME/lib

      Build from source and copy the include and libraries into ORT_HOME

      On Windows

      build.bat --config RelWithDebInfo --build_shared_lib --skip_tests --parallel [--use_cuda]
      copy include\onnxruntime\core\session\onnxruntime_c_api.h $ORT_HOME\include
      copy build\Windows\RelWithDebInfo\RelWithDebInfo\*.dll $ORT_HOME\lib

      On Linux

      ./ --build_shared_lib --skip_tests --parallel [--use_cuda]
      cp include/onnxruntime/core/session/onnxruntime_c_api.h $ORT_HOME/include
      cp build/Linux/RelWithDebInfo/libonnxruntime*.so* $ORT_HOME/lib
  3. Build onnxruntime-genai

    If you are building for CUDA, add the cuda_home argument.

    cd ..
    python [--cuda_home <path_to_cuda_home>]
  4. Install Python wheel

    cd build/wheel
    pip install *.whl

Model download and export

ONNX models are run from a local folder, via a string supplied to the Model() method.

To source microsoft/phi-2 optimized for your target, download and run the following script. You will need to be logged into Hugging Face via the CLI to run the script.

Install model builder dependencies.

pip install numpy
pip install transformers
pip install torch
pip install onnx
pip install onnxruntime

Export int4 CPU version

huggingface-cli login --token <your HuggingFace token>
python -m onnxruntime_genai.models.builder -m microsoft/phi-2 -p int4 -e cpu -o <model folder>


This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact with any additional questions or comments.


This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.

Product Compatible and additional computed target framework versions.
.NET net5.0 was computed.  net5.0-windows was computed.  net6.0 was computed.  net6.0-android was computed.  net6.0-ios was computed.  net6.0-maccatalyst was computed.  net6.0-macos was computed.  net6.0-tvos was computed.  net6.0-windows was computed.  net7.0 was computed.  net7.0-android was computed.  net7.0-ios was computed.  net7.0-maccatalyst was computed.  net7.0-macos was computed.  net7.0-tvos was computed.  net7.0-windows was computed.  net8.0 was computed.  net8.0-android was computed.  net8.0-browser was computed.  net8.0-ios was computed.  net8.0-maccatalyst was computed.  net8.0-macos was computed.  net8.0-tvos was computed.  net8.0-windows was computed. 
.NET Core netcoreapp3.0 was computed.  netcoreapp3.1 was computed. 
.NET Standard netstandard2.1 is compatible. 
MonoAndroid monoandroid was computed. 
MonoMac monomac was computed. 
MonoTouch monotouch was computed. 
native native is compatible. 
Tizen tizen60 was computed. 
Xamarin.iOS xamarinios was computed. 
Xamarin.Mac xamarinmac was computed. 
Xamarin.TVOS xamarintvos was computed. 
Xamarin.WatchOS xamarinwatchos was computed. 
Compatible target framework(s)
Included target framework(s) (in package)
Learn more about Target Frameworks and .NET Standard.

NuGet packages (1)

Showing the top 1 NuGet packages that depend on Microsoft.ML.OnnxRuntimeGenAI.Cuda:

Package Downloads

Semantic Kernel connector for Microsoft.ML.OnnxRuntimeGenAI.

GitHub repositories

This package is not used by any popular GitHub repositories.

Version Downloads Last updated
0.3.0-rc2 771 5/29/2024
0.3.0-rc1 488 5/22/2024
0.2.0 1,099 5/20/2024
0.2.0-rc7 788 5/14/2024
0.2.0-rc6 772 5/4/2024
0.2.0-rc4 1,211 4/25/2024
0.2.0-rc3 188 4/24/2024
0.1.0 274 4/8/2024
0.1.0-rc4 222 3/27/2024

Introducing the ONNX Runtime GenAI Library.