unifyai 0.5.0

A Python package for interacting with the Unify API

Unify Python API Library

The Unify Python Package provides access to the Unify REST API, allowing you to query Large Language Models (LLMs) from any Python 3.7.1+ application. It includes Synchronous and Asynchronous clients with Streaming responses support.

Just like the REST API, you can:

• 🔑 Use any endpoint with one key: Access all LLMs at any provider with just one Unify API Key.

• 🚀 Route to the best endpoint: Each prompt is sent to the endpoint that will yield the best performance for your target metric, including high-throughput, low cost or low latency. See the routing section to learn more about this!

Installation

You can use pip to install the package as follows:

pip install unifyai

Basic Usage

import os
from unifyai import Unify
unify = Unify(
    # This is the default and optional to include.
    api_key=os.environ.get("UNIFY_KEY")
    model="llama-2-13b-chat@anyscale"
)
response = unify.generate(messages="Hello Llama! Who was Isaac Newton?")

Here, response is a string containing the model's output.

You can influence the model's persona using the system_prompt argument in the .generate function:

response = unify.generate(messages="Hello Llama! Who was Isaac Newton?", system_prompt="You should always talk in rhymes")

If you want change the model, you can do so by updating the .model attribute of the client:

client.model = "mistral-7b-instruct-v0.1@deepinfra"

Supported Models

The list of supported models and providers is available in the platform.

API Key

You can get an API Key from the Unify console

[!NOTE] You can provide an api_key keyword argument, but we recommend using python-dotenv to add UNIFY_KEY="My API Key" to your .env file so that your API Key is not stored in source control.

Sending multiple messages

When a string is passed to the messages argument, it is assumed to be the user prompt. However, you can also pass a list of dictionaries containing the message history between the user and the assistant, as shown below:

messages=[
   {"role": "user", "content": "Who won the world series in 2020?"},
   {"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."},
   {"role": "user", "content": "Where was it played?"}
]
res = unify.generate(messages=messages)

Asynchronous Usage

For optimal performance in handling multiple user requests simultaneously, such as in a chatbot application, processing them asynchronously is recommended. To use the AsyncUnify client, simply import AsyncUnify instead of Unify and use await with the .generate function.

from unifyai import AsyncUnify
import os
import asyncio
async_unify = AsyncUnify(
   # This is the default and optional to include.
   api_key=os.environ.get("UNIFY_KEY")
   model="llama-2-13b-chat@anyscale"
)

async def main():
   responses = await async_unify.generate(messages="Hello Llama! Who was Isaac Newton?")

asyncio.run(main())

Functionality wise, the Async and Sync clients are identical.

Streaming Responses

You can enable streaming responses by setting stream=True in the .generate function.

import os
from unifyai import Unify
unify = Unify(
    # This is the default and optional to include.
    api_key=os.environ.get("UNIFY_KEY")
    model="llama-2-13b-chat@anyscale"
)
stream = unify.generate(messages="Hello Llama! Who was Isaac Newton?")
for chunk in stream:
    print(chunk, end="")

It works in exactly the same way with Async clients.

from unifyai import AsyncUnify
import os
import asyncio
async_unify = AsyncUnify(
   # This is the default and optional to include.
   api_key=os.environ.get("UNIFY_KEY")
   model="llama-2-13b-chat@anyscale"
)

async def main():
   async_stream = await async_unify.generate(messages="Hello Llama! Who was Isaac Newton?")
   async for chunk in async_stream:
       print(chunk, end="")

asyncio.run(main())

Get Current Credit Balance

You can use the .get_credit_balance method to the credit balance for the authenticated account as follows:

credits = unify.get_credit_balance()

Dynamic Routing

As evidenced by our benchmarks, the optimal provider for each model varies by geographic location and time of day due to fluctuating API performances. With our dynamic routing, we automatically direct your requests to the "top-performing provider" at that moment. To enable this feature, simply replace your query's provider with one of the available routing modes. As an example, you can query the llama-2-7b-chat endpoint to get the provider with the lowest input-cost as follows:

import os
from unifyai import Unify
unify = Unify(
    # This is the default and optional to include.
    api_key=os.environ.get("UNIFY_KEY")
    model="llama-2-13b-chat@lowest-input-cost"
)
response = unify.generate(messages="Hello Llama! Who was Isaac Newton?")

You can see the provider chosen by printing the .provider attribute of the client:

print(unify.provider)

Dynamic routing works with both Synchronous and Asynchronous clients. For more information on Dynamic Routing, check our documentation.