Async version in aisuite.AsyncClient class

[kapulkin]

Async LLM calls support proposal

Description

I added aisuite.AsyncClient interface with async version of llm calls and implemented support for a few providers: OpenAI, Anthropic, Mistral, Fireworks. Async tests for supported providers are written.

Demand

Async calls are necessary for industrial development and have huge adoption in python libraries like FastAPI. It is easy and popular way to run llm calls in parallel with asyncio.gather() function.

The PR addresses request #61

Changes Made

Class aisuite.AsyncClient is added. The new method def chat_completions_create_async(self, model, messages) is added into Provider class. Default implementation throws NotImplementedError.

Checklist

• Tests are written
• Pre-commit hooks pass

[kapulkin]

@ksolo @foxty @jeffxtang @joaomdmoura can you, please, review the PR?

[kapulkin]

I synchronized async client version by adding tool and thinking support.

@rohitprasad15, review, please, the PR.

[antoniomuzzolini]

Hey, found a bug! If you try to set providers

import aisuite_async as ai
providers = {"openai": {"api_key": custom_api_key}}
llm = ai.AsyncClient(provider_configs=providers)

the provider is istantiated by Client, thus not async, and then throws error 'cause chat_completions_create_async method doesn't exists.

I think that overriding the _initialize_providers method setting is_async=True should be enough.

[kapulkin]

@antoniomuzzolini Thank you for the report!

I made more general update by introducing BaseClient class which avoids copying and pasting. Check, please, the new version.