The goal of the Banana CLI is to bring an npm-like experience to ML development loops.

This version includes:

• banana init to create a project with boilerplate
• banana install to install packages from the requirements.txt
• banana dev to run a dev server with hot-reload (beta)

Here's a demo video

1. Install the CLI with pip

pip3 install banana-cli

2. Create a new project directory with

banana init my-app
cd my-app

3. Start the dev server

python3 app.py

4. Call your API (from a separate terminal)

curl -X POST -H "Content-Type: application/json" -d '{"prompt": "Hello I am a [MASK] model."}' http://localhost:8000/

You may try the interractive dev server, in beta, with

banana dev

The interactive dev server works like a react, next, or nodemon server: it selectively hot reloads components when you save changes to different parts of your app.py file.

The init() function is ran on startup and for every change to init().

The handler() function is ran on every change to handler(), without needing to wait for a long init()

4. Try changing the handler, see what happens!
5. Try changing the init, see what happens!

• banana dev --auto-compat=True to make your GPU code compatible with a CPU machine, by ignoring to("cuda") calls

• Lock in a stable interface
• Add the following commands
  • banana build -> verify production build
  • banana test -> unit test against local test cases
  • banana deploy -> manually deploy from CLI
  • banana deploy --canary --ttl=10 -> run a temporary deployment to Banana's cluster for on-GPU testin
• Port to Rust. Why Rust?