Skip to content

Environment variables

Each API has a number of environment variables that are either required or optional to be set in order to query the model. We recommend setting these environment variables in a .env file in the root of the project directory. When you run the prompto_run_experiment or prompto_run_pipeline commands, the library will look for an .env file in the current directory and we then use python-dotenv to load the environment variables into the Python environment.

An .env file would look something like: 

OPENAI_API_KEY=<YOUR-OPENAI-KEY>
AZURE_OPENAI_API_KEY=<YOUR-AZURE-OPENAI-KEY>
OLLAMA_API_ENDPOINT=<YOUR-OLLAMA_API-ENDPOINT>

This allows you to not necessarily need to load them into the global environment in your terminal each time you run an experiment.

Alternatively, you can set the environment variables in your terminal using export ENVIRONMENT_VARIABLE=value or having another bash script to set these without a .env file * this file would look similar to the example .env file, but with export in front of each line so that it is loaded into the global environment.

We list the environment variables for each API in the their respective documentation pages. See the models doc for a list of all the models and their respective documentation pages.

Model-specific environment variables

There may be some experiments where you wish to query different models from the same API type (e.g. querying different deployments of models from an Azure OpenAI API). In such cases, you can specify the model name in the model_name key of the prompt_dict.

There may be some cases where the environment variables needed for different models are different, for example if you are querying models which have different API keys (in Azure OpenAI, you may have different models that live on different subscriptions). In such cases, you can set model-specific environment variables by appending the model name to the environment variable name (with a underscore between the environment variable and the model name): ENVIRONMENT_VARIABLE -> ENVIRONMENT_VARIABLE_model_name. Note that the model name might have to be formatted to be a valid environment variable name (e.g. replacing ("-", "/", ".", " ") with underscores _).

For example, if you have two models in Azure OpenAI, model1 and model2, you can set the environment variables AZURE_OPENAI_API_KEY_model1 and AZURE_OPENAI_API_KEY_model2 with the respective API keys for each model.

The base environment variable name (e.g. AZURE_OPENAI_API_KEY) is used as a default environment variable if the model-specific environment variable is not set. They are used either in the case where the model name is not specified in the prompt_dict or if the model-specific environment variable is not set.

To clarify, the order of precedence for the API key is as follows:

  1. If model_name is specified in the prompt_dict, the model-specific environment variable is used
  2. If model_name is specified in the prompt_dict but the model-specific environment variable is not set, the default environment variable is used
  3. If model_name is not specified in the prompt_dict, the default environment variable is used
  4. If neither the model-specific environment variable nor the default environment variable is set, an error is raised