Quick Start

Start using OpenPO in your code within a minute.

Set your API Key

Configure your API key of a model of your choice as an environment variable.

export HF_API_KEY=<your-huggingface-api-key>
export OPENROUTER_API_KEY=<your-openrouter-api-key>

Using Hugging Face

OpenPO client uses Hugging Face's InferenceClient to call models available on Huggingface Model Hub.

Inference API

import os
from openpo.client import OpenPO

client = OpenPO(api_key=os.getenv("HF_API_KEY"))

response = client.chat.completions.create_preference(
    model="Qwen/Qwen2.5-Coder-32B-Instruct",
    messages=[
        {"role": "system", "content": PROMPT},
        {"role": "system", "content": MESSAGE},
    ],
    diff_frequency=0.5, # generate pairwise responses 50% of the time
)

print(res.choices[0].message.content)

Inference Endpoint

To call models that are deployed to Hugging Face Inference Endpoint, simply pass in the private url to the model parameter.

client = OpenPO(api_key=os.getenv("HF_API_KEY"))

response = client.chat.completions.create_preference(
    model="<your-inference-endpoint-url>",
    messages=[
        {"role": "system", "content": PROMPT},
        {"role": "system", "content": MESSAGE},
    ],
    diff_frequency=0.5, # generate pairwise responses 50% of the time
)

Using OpenRouter

OpenPO supports OpenRouter by making requests to its endpoint.

client = OpenPO(
    base_url="https://openrouter.ai/api/v1/chat/completions",
    api_key=os.getenv("OPENROUTER_API_KEY")
)

response = client,.chat.completions.create_preference(
    model="qwen/qwen-2.5-coder-32b-instruct",
    messages=[
        {"role": "system", "content": PROMPT},
        {"role": "system", "content": MESSAGE},
    ],
    diff_frequency=0.5,
)