ONNX Runtime generate() API

Run Llama, Phi, Gemma, Mistral with ONNX Runtime.

This API gives you an easy, flexible and performant way of running LLMs on device.

It implements the generative AI loop for ONNX models, including pre and post processing, inference with ONNX Runtime, logits processing, search and sampling, and KV cache management.

You can call a high level generate() method to generate all of the output at once, or stream the output one token at a time.

See documentation at https://onnxruntime.ai/docs/genai.

Support matrix	Supported now	Under development	On the roadmap
Model architectures	Gemma Llama * Mistral + Phi (language + vision) Qwen Nemotron	Whisper	Stable diffusion
API	Python C# C/C++ Java ^	Objective-C
Platform	Linux Windows Mac ^ Android ^		iOS
Architecture	x86 x64 Arm64 ~
Hardware Acceleration	CUDA DirectML	QNN ROCm	OpenVINO
Features		Interactive decoding Customization (fine-tuning)	Speculative decoding

* The Llama model architecture supports similar model families such as CodeLlama, Vicuna, Yi, and more.

+ The Mistral model architecture supports similar model families such as Zephyr.

^ Requires build from source

~ Windows builds available, requires build from source for other platforms

Installation

See https://onnxruntime.ai/docs/genai/howto/install

Sample code for Phi-3 in Python

Download the model

huggingface-cli download microsoft/Phi-3-mini-4k-instruct-onnx --include cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/* --local-dir .

Install the API

pip install numpy
pip install --pre onnxruntime-genai

Run the model

import onnxruntime_genai as og

model = og.Model('cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4')
tokenizer = og.Tokenizer(model)
tokenizer_stream = tokenizer.create_stream()
 
# Set the max length to something sensible by default,
# since otherwise it will be set to the entire context length
search_options = {}
search_options['max_length'] = 2048

chat_template = '<|user|>\n{input} <|end|>\n<|assistant|>'

text = input("Input: ")
if not text:
   print("Error, input cannot be empty")
   exit

prompt = f'{chat_template.format(input=text)}'

input_tokens = tokenizer.encode(prompt)

params = og.GeneratorParams(model)
params.set_search_options(**search_options)
params.input_ids = input_tokens
generator = og.Generator(model, params)

print("Output: ", end='', flush=True)

try:
   while not generator.is_done():
     generator.compute_logits()
     generator.generate_next_token()

     new_token = generator.get_next_tokens()[0]
     print(tokenizer_stream.decode(new_token), end='', flush=True)
except KeyboardInterrupt:
    print("  --control+c pressed, aborting generation--")

print()
del generator

Roadmap

See the Discussions to request new features and up-vote existing requests.

Contributing

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 https://cla.opensource.microsoft.com.

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 opencode@microsoft.com with any additional questions or comments.

Trademarks