Settings¶

Marvin makes use of Pydantic's BaseSettings for configuration throughout the package.

Environment Variables¶

All settings are configurable via environment variables like MARVIN_<setting name>.

For example, in an .env file or in your shell config file you might have:

MARVIN_LOG_LEVEL=DEBUG
MARVIN_LLM_MODEL=openai/gpt-4
MARVIN_LLM_TEMPERATURE=0

Runtime Settings¶

A runtime settings object is accessible via marvin.settings and can be used to access or update settings throughout the package.

For example, to access or change the LLM model used by Marvin at runtime:

import marvin

marvin.settings.llm_model
# 'openai/gpt-4'

marvin.settings.llm_model = 'openai/gpt-3.5-turbo'

marvin.settings.llm_model
# 'openai/gpt-3.5-turbo'

LLM Providers¶

Marvin supports multiple LLM providers, including OpenAI and Anthropic. After configuring your credentials appropriately, you can use any supported model by setting marvin.settings.llm_model appropriately.

Valid llm_model settings are strings with the form "{provider_key}/{model_name}". For example, "openai/gpt-3.5-turbo", anthropic/claude-2, or azure_openai/gpt-4.

Provider	Provider Key	Models	Notes
OpenAI	`openai`	`gpt-3.5-turbo`, `gpt-4` (default), or any other compatible model	Marvin is generally tested and optimized with this provider.
Anthropic	`anthropic`	`claude-2`, `claude-instant-1` or any other compatible model	Available in Marvin 1.1
Azure OpenAI Service	`azure_openai`	`gpt-3.5-turbo`, `gpt-4`, or any other compatible model	The Azure OpenAI Service shares all the same configuration options as the OpenAI models, as well as a few additional ones. Available in Marvin 1.1.

LLM Configuration¶

To configure LLM models globally, you can adjust the following settings. Note that these become the default settings for all models, but you can always set these on a per-model or per-component basis.

Setting	Env Variable	Runtime Variable	Default	Notes
LLM model	`MARVIN_LLM_MODEL`	`marvin.settings.llm_model`	`openai/gpt-3.5-turbo`	Set the model as `{provider}/{model}`. Defaults to OpenAI's GPT-3.5 model.
Temperature	`MARVIN_LLM_TEMPERATURE`	`marvin.settings.llm_temperature`	0.8
Max tokens	`MARVIN_LLM_MAX_TOKENS`	`marvin.settings.llm_max_tokens`	1500	The maximum number of tokens in a model completion
Timeout	`MARVIN_LLM_REQUEST_TIMEOUT_SECONDS`	`marvin.settings.llm_request_timeout_seconds`	600.0