Secrets - Function

Inject your Environment Variables and API Keys.

Function provides functionality for injecting environment variables into predictors while they are being provisioned. These are primarily used to pass secrets to the predictor, including API keys and other secure information.

Function stores all environment variables with encryption. As such, secrets and environment variables are the equivalent.

In this guide, let's make a predictor uses OpenAI's managed Whisper service to transcribe an audio file:

Implementing the Predictor

Create a transcribe.ipynb notebook and add the following cell to install a few dependencies we'll need:

# Install OpenAI library
%pip install openai

Next, add a new code cell and paste the following implementation:

import openai
from pathlib import Path

def predict (audio: Path) -> str:
    # Make sure that the audio file has a ".wav" extension
    audio = audio.rename(audio.with_suffix(f".wav"))
    # Transcribe
    with open(audio, "rb") as f:
        transcript = openai.Audio.transcribe("whisper-1", f)
    # Return
    return transcript["text"]

This predictor accepts an audio path pointing to an audio file on the system. It then transcribes the file using OpenAI's hosted Whisper service, then returns the transcribed text.

Function serializes any audio input value to a Path which points to the audio file. See more on Value →

Using Secrets in Code

We have to specify our OpenAI API key to use the Whisper service in our code. In Function, environment variables are accessed through the os.environ dictionary. Let's now update our code to set the API key:

import openai
from os import environ
from pathlib import Path

# Specify our OpenAI API key
openai.api_key = environ.get("OPENAI_API_KEY")

def predict (audio: Path) -> str:
    # Make sure that the audio file has a ".wav" extension
    audio = audio.rename(audio.with_suffix(f".wav"))
    # Transcribe
    with open(audio, "rb") as f:
        transcript = openai.Audio.transcribe("whisper-1", f)
    # Return
    return transcript["text"]

During provisioning, Function injects environment variables to your predictor container, making them globally accessible.

We technically don't need the added lines above because the OpenAI library automatically finds the OPENAI_API_KEY ↗ env.

Defining Secrets

Function provides two ways to define environment variables to predictors:

Global Secrets

Global secrets are available to all predictors under a given account. To define a global secret, use the Function CLI:

# Open a terminal and run the following command:
fxn env create OPENAI_API_KEY 1234abcd

You can use the --organization flag to create a global secret for an organization account.

Once a secret is created, its value can only be viewed on fxn.ai ↗

You can also define global secrets on fxn.ai ↗

specifying global secrets on Function

Predictor Secrets

As the name implies, predictor secrets are specific to a given predictor. To define a predictor secret, use the --env flag when creating the predictor in the Function CLI:

# Open a terminal and run this command to create our predictor
fxn create @username/transcribe transcribe.ipynb --env OPENAI_API_KEY=1234abcd

Replace @username with your Function username.

When defining predictor secrets, you must use the following format: --env NAME=VALUE.

Making a Prediction

If you haven't already, head over to OpenAI ↗ to generate a new API key. Then define a new global secret:

# Open a terminal and run the following command:
fxn env create OPENAI_API_KEY <YOUR OPENAI API KEY>

We recommend creating API keys as global secrets on Function.

Now, let's create the predictor:

# Create the predictor
fxn create @username/transcribe transcribe.ipynb

Replace @username with your Function username.

Once the predictor has been provisioned, head over to voice-recorder-online.com ↗ to record your voice to a waveform audio file.

Function makes no guarantees about the functionality or security of voice-recorder-online.com.

Next, open a terminal and run the following command:

# Test out our transcription predictor
fxn predict @username/transcribe --audio @record.wav

After a few moments, you should see the transcription:

making a prediction with secrets