In Your First Model, we walked through how to deploy a basic model to Baseten. If you are trying to rapidly make changes and iterate on your model, you’ll notice that there is quite a bit of time between running truss push and when the changes are reflected on Baseten.

Also, a lot of models require special hardware that you may not immediately have access to.

To solve this problem, we have a feature called Truss Watch, that allows you to live reload your model as you work.

​

Truss Watch

To make use of truss watch, start by deploying your model:

$ truss push

By default, this will deploy a “development” version of your model. This means that the model has a live reload server attached to it and supports hot reloading. To get the hot reload loop working, simply run truss watch afterwards:

$ truss watch

Now, if you make changes to your model, you’ll see them reflected in the model logs!

You can now happily iterate on your model without having to go through the entire build & deploy loop between each change.

​

Ready for Production?

Once you’ve iterated on your model, and you’re ready to deploy it to production, you can use the truss push --publish command. This will deploy a “published” version of your model

truss push --publish

Note that development models have slightly worse performance, and have more limited scaling properites, so it’s highly recommend to not use these for any production use-case.