vendi-sdk 0.2.3

Vendi SDK

---

Documentation 📖: https://docs.vendi-ai.com

Source Code 💻 : https://github.com/vendi-ai/vendi-python

---

Vendi Python API Library

The Vendi python library provides convenient access to the Vendi REST API from any Python 3.7+ application. The library includes all access to the Vendi API, including the ability to create, update, and delete resources.

Installation

Install the latest version of the library with pip:

pip install vendi

Usage

The library needs to be configured with your account's API key which is available in the Vendi web interface.

from vendi import Vendi

client = Vendi(api_key="YOUR_API_KEY") 

Or you can set the VENDI_API_KEY environment variable and the library will automatically use it:

export VENDI_API_KEY="YOUR_API_KEY"

Inference

The library provides a convenient way to make inference requests to your models.

from vendi import Vendi

client = Vendi(api_key="YOUR_API_KEY")

chat_completion = client.completions.create(
    model_id="vendi/mistral-7b-instruct-v2",
    messages=[
        {
            "role": "user",
            "content": "Hello, I'm looking for a new job. Write me a resume."
        }
    ],
    max_tokens=5,
    temperature=0.5,
    top_p=1,
    frequency_penalty=0.5,
    presence_penalty=0.5,
    stop=["\n"]
)

print(chat_completion.provider_response.choices[0].text)

The inference endpoints are OpenAI compatible, so you can use the same parameters as the OpenAI API or even the OpenAI Python library.

Datasets

The library provides a convenient way to upload and download datasets from your account.

from vendi import Vendi

client = Vendi(api_key="YOUR_API_KEY")

dataset = client.datasets.upload(
    name="My Dataset",
    data=[
        {
            "input": "The bill is $5.",
            "output": "5$"
        },
        {
            "name": "They gave us a nice $10 bill.",
            "output": "10$"
        }
    ]
)

You can also download datasets from your account.

from vendi import Vendi

client = Vendi(api_key="YOUR_API_KEY")

dataset = client.datasets.get("YOUR_DATASET_ID")

data = dataset.load_data()

Finetune

The library provides a convenient way to finetune models from your account.

from vendi import Vendi

client = Vendi(
    api_key="my-api-key"
)

dataset = client.datasets.upload(
    name="my-dataset",
    data_path="conversation.jsonl"
)

finetune_job = client.finetune.run(
    run_name="my-first-finetune",
    model_description="My first finetune",
    model_name="vendi/mistralai/Mistral-7B-Instruct-v0.2",
    dataset_id=dataset.id
)

print(finetune_job)

The example above will create a finetune job that will run on the Vendi platform. You can monitor the progress of the finetune job in the Vendi web interface.

To finetune other models rather than mistralai, you can use the client.finetune.available_models() method.

Models

The models are the finetuned models created out of the finetune jobs. You can list all the models in your account with the client.models.list() method.

from vendi import Vendi

client = Vendi(
    api_key="my-api-key"
)

client.models.list()

Inference on finetuned model

Every finetuned model can be used for inference. You can use the client.completions.create() method to make inference requests to your models just like any other pre-trained models.

from vendi import Vendi

client = Vendi(
    api_key="my-api-key"
)

client.completions.create(
    model_id="<account-name>/<finetuned-model-name>",
    messages=[
        {
            "role": "user",
            "content": "Hello, I'm looking for a new job. Write me a resume."
        }
    ],
    max_tokens=256,
    temperature=0.5,
    top_p=1,
    frequency_penalty=0.5,
    presence_penalty=0.5,
    stop=["\n"]
)

📚 For a thorough introduction, features, and best practices, explore our official documentation and quickstart.

Contributing 💼

Join our mission to make declarative AI even better together! We welcome contributions from everyone. Please read our contributing guide to get started.