/serverless-template-gpt

Primary LanguagePythonMIT LicenseMIT

🍌 Banana Serverless

This repo gives a basic framework for serving ML models in production using simple HTTP servers.

Quickstart:

The repo is already set up to run a basic HuggingFace BERT model.

  1. Run pip3 install -r requirements.txt to download dependencies.
  2. Run python3 server.py to start the server.
  3. Run python3 test.py in a different terminal session to test against it.

Make it your own:

  1. Edit app.py to load and run your model.
  2. Make sure to test with test.py!

if deploying using Docker:

  1. Edit download.py (or the Dockerfile itself) with scripts download your custom model weights at build time.

Move to prod:

At this point, you have a functioning http server for your ML model. You can use it as is, or package it up with our provided Dockerfile and deploy it to your favorite container hosting provider!

If Banana is your favorite GPU hosting provider (and we sure hope it is), read on!

🍌

Deploy to Banana Serverless:

Four steps:

  1. Create your own copy of this template repo. Either:
  • Click "Fork" (creates a public repo)
  • Click "Use this Template" (creates a private or public repo)
  • Create your own repo and copy the template files into it

  1. Install the Banana Github App to your new repo.

  2. Get your Banana API Key by logging in here.

  3. Email us at onboarding@banana.dev with the following message:

Hello, I'd like to be onboarded to serverless.
My github username is: YOUR_GITHUB_USERNAME
My Banana API Key is: YOUR_API_KEY
My preferred billing email is: YOU@EMAIL.COM

Your github username, banana api key, and email are required for us to authorize you into the system. We will reply and confirm when you're added.

From then onward, any pushes to the main branch trigger Banana to build and deploy your server, using the Dockerfile. Throughout the build we'll sprinkle in some secret sauce to make your server extra snappy 🔥

It'll then be deployed on our Serverless GPU cluster and callable with any of our serverside SDKs:

Use Banana for scale!