Providers

OpenAI¶

Marvin supports OpenAI's GPT-3.5 and GPT-4 models, and uses the openai/gpt-4 model by default. In order to use the OpenAI API, you must provide an API key.

Configuration¶

To use OpenAI models, you can set the following configuration options:

Setting	Env Variable	Runtime Variable	Required?	Notes
API key	`MARVIN_OPENAI_API_KEY`	`marvin.settings.openai.api_key`	✅

Using the Azure OpenAI Service

To use the Azure OpenAI Service, configure it explicitly.

Getting an API key¶

To obtain an OpenAI API key, follow these steps:

Log in to your OpenAI account (sign up if you don't have one)
Go to the "API Keys" page under your account settings.
Click "Create new secret key." A new API key will be generated. Make sure to copy the key to your clipboard, as you will not be able to see it again.

Setting the API key¶

You can set your API key at runtime like this:

import marvin

# Marvin 1.1+
marvin.settings.openai.api_key = YOUR_API_KEY

# Marvin 1.0
marvin.settings.openai_api_key = YOUR_API_KEY

However, it is preferable to pass sensitive settings as an environment variable: MARVIN_OPENAI_API_KEY.

To set your OpenAI API key as an environment variable, open your terminal and run the following command, replacing with the actual key:

export MARVIN_OPENAI_API_KEY=<your API key>

This will set the key for the duration of your terminal session. To set it more permanently, configure your terminal or its respective env files.

Using OpenAI standard API key locations

For convenience, Marvin will respect the OPENAI_API_KEY environment variable or a key manually set as openai.api_key as long as no Marvin-specific keys were also provided.

Using a model¶

Once your API key is set, you can use any valid OpenAI model by providing it as Marvin's llm_model setting:

import marvin

marvin.settings.llm_model = 'openai/gpt-4-0613'

Anthropic¶

Available in Marvin 1.1

Marvin supports Anthropic's Claude 1 and Claude 2 models. In order to use the Anthropic API, you must provide an API key.

Installing the Anthropic provider

To use the Anthropic provider, you must have the anthropic Python client installed. You can do this by installing Marvin as pip install "marvin[anthropic]"

Anthropic is not optimized for calling functions

Anthropic's models are not fine-tuned for calling functions or generating structured outputs. Therefore, Marvin adds a significant number of additional instructions to get Anthropic models to mimic this behavior. Empirically, this works very well for most Marvin components, including functions, models, and classifiers. However, it may not perform as well for more complex AI Applications.

Configuration¶

To use Anthropic models, you can set the following configuration options:

Setting	Env Variable	Runtime Variable	Required?	Notes
API key	`MARVIN_ANTHROPIC_API_KEY`	`marvin.settings.anthropic.api_key`	✅

Getting an API key¶

To obtain an Anthropic API key, follow these steps:

Log in to your Anthropic account (sign up if you don't have one)
Go to the "API Keys" page under your account settings.
Click "Create Key." A new API key will be generated. Make sure to copy the key to your clipboard, as you will not be able to see it again.

Setting the API key¶

You can set your API key at runtime like this:

import marvin

marvin.settings.anthropic.api_key = YOUR_API_KEY

However, it is preferable to pass sensitive settings as an environment variable: MARVIN_ANTHROPIC_API_KEY.

To set your Anthropic API key as an environment variable, open your terminal and run the following command, replacing with the actual key:

export MARVIN_ANTHROPIC_API_KEY=<your API key>

This will set the key for the duration of your terminal session. To set it more permanently, configure your terminal or its respective env files.

Using a model¶

Once your API key is set, you can use any valid Anthropic model by providing it as Marvin's llm_model setting:

import marvin

marvin.settings.llm_model = 'claude-2'

Marvin will automatically recognize that the claude-* family of models use the Anthropic provider. To indicate a provider explicitly, prefix the model name with anthropic/. For example: marvin.settings.llm_model = 'anthropic/claude-2'.

Azure OpenAI Service¶

Available in Marvin 1.1

Marvin supports Azure's OpenAI service. In order to use the Azure service, you must provide an API key.

Configuration¶

To use the Azure OpenAI service, you can set the following configuration options:

Setting	Env Variable	Runtime Variable	Required?	Notes
API key	`MARVIN_AZURE_OPENAI_API_KEY`	`marvin.settings.azure_openai.api_key`	✅
API base	`MARVIN_AZURE_OPENAI_API_BASE`	`marvin.settings.azure_openai.api_base`	✅	The API endpoint; this should have the form `https://YOUR_RESOURCE_NAME.openai.azure.com`
Deployment name	`MARVIN_AZURE_OPENAI_DEPLOYMENT_NAME`	`marvin.settings.azure_openai.deployment_name`	✅
API type	`MARVIN_AZURE_OPENAI_API_TYPE`	`marvin.settings.azure_openai.api_type`		Either `azure` (the default) or `azure_ad` (to use Microsoft Active Directory to authenticate to your Azure endpoint).

Using a model¶

Once the provider is configured, you can use any Azure OpenAI model by providing it as Marvin's llm_model setting:

import marvin

marvin.settings.llm_model = 'azure_openai/gpt-35-turbo'