martian-adapters

LLM Adapters Package Documentation

List of currently supported models

Overview

The Adapters package facilitates communication between different language model APIs by providing a unified interface for interaction. This ensures ease of use and flexibility in integrating multiple models from various providers.

The package can be installed an used via pip:

pip install martian-adapters

Getting Started

Prerequisites

• Python version: 3.11.10
• Poetry

Installation

poetry install
poetry run pre-commit install

Setting Up Pre-commit

To run pre-commit manually:

poetry run pre-commit run --all-files

Semantic Versioning

For versioning we follow Semantic Versioning

Environment Configuration

The package requires certain environment variables to be set by the users:

• Copy .env-example to .env and populate it with appropriate values.

Running Tests

poetry run pytest

Quickstart

from adapters import AdapterFactory

# First component in model path is Provider, then Vendor, and last model name itself
adapter = AdapterFactory.get_adapter_by_path("openai/openai/gpt-4o-mini")

adapter.execute_sync(
    [
        {"role": "system", "content": "You are a helpful assistant."},
        {
            "role": "user",
            "content": "Write a haiku about recursion in programming.",
        },
    ]
)

Adapter paths follows the format provider/vendor/model_name. Use AdapterFactory.get_supported_models() to retrieve all supported models. For a given model, model.get_path() returns the adapter path.

Contributing

Adding New Models

• Existing Providers: Add new models to the MODELS array if the provider is already supported.

• New Providers:

  • If the provider follows the OpenAI format, model integration is straightforward. See the "Fireworks" provider class as an example.
  • For providers with different schemas, see the "Anthropic" provider class for guidance.

Development Steps

• Add the Provider and Model: Update provider_adapters/__init__.py and test files accordingly.

• Write Tests: Add tests in the relevant directories. Use @pytest.mark.vcr for tests making network requests.

• Run Tests:

  poetry run pytest
  

• Check-in Cassette Files: Include any new cassette YAML files in your commit.

• Send a Pull Request: Ensure all tests pass before requesting a review.

Re-creating Cassette Files

Use the poetry run pytest --record-mode=rewrite option with pytest to update cassette files.

Additional Notes

Some models may only be accessible from specific locations (e.g., the U.S.). In such cases, running tests might require access to a U.S.-based server.

This documentation provides a streamlined approach to using and contributing to the Adapters package, emphasizing practical steps and clear examples.

Misclenous

HTTP Client configuration

To optimize throughput and performance, we provide options to configure HTTP networking parameters:

ADAPTERS_MAX_KEEPALIVE_CONNECTIONS_PER_PROCESS = 100
ADAPTERS_MAX_CONNECTIONS_PER_PROCESS = 1000
ADAPTERS_HTTP_CONNECT_TIMEOUT = 5
ADAPTERS_HTTP_TIMEOUT = 600

Base URL overriding

For stress testing or other purposes, you can override all base URLs by setting the following in your .env file:

_ADAPTERS_OVERRIDE_ALL_BASE_URLS_ = "https://new-base-url.com/api"

This setting ensures that all LLM API calls will route to the specified new base URL.